Energy Systems

Atom Mirakyan
Roland De Guio

Three Domain
Modelling and
Uncertainty
Analysis
Applications in Long Range
Infrastructure Planning
Energy Systems

Series editor
Panos M. Pardalos, Gainesville, USA
More information about this series at http://www.springer.com/series/8368
Atom Mirakyan · Roland De Guio

Three Domain Modelling

and Uncertainty Analysis
Applications in Long Range Infrastructure
Planning

13
Atom Mirakyan Roland De Guio
European Institute for Energy INSA de Strasbourg
Research (EIFER) Strasbourg
Karlsruhe France
Germany

ISSN 1867-8998 ISSN 1867-9005 (electronic)

Energy Systems
ISBN 978-3-319-19571-1 ISBN 978-3-319-19572-8 (eBook)
DOI 10.1007/978-3-319-19572-8

Library of Congress Control Number: 2015940428

Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
For my parents convincing me to accept
some uncertainties and for my wife
Preface

This book is the synthesis of the results of several research activities during our
previous 10 years of experience in energy planning and modelling. While some
elements of this work have been published in a Ph.D. thesis, journals and pro-
ceedings of international conferences, until now there had been no single resource
that described all of these works in a unified view. These scientific and practical
experiences are the basis of this book. However, new observations and experiences
have also been included, and the discussion is organized for a broad audience.
The central problem discussed here is that the local infrastructure, e.g., energy
systems, has numerous and diverse subsystems, nonlinear interactions, multi-
ple scales and heterogeneity. Planning and modelling such systems over the long
range is a complex task connected with different uncertainties. In addressing this
problem, the reader will find answers to several questions such as:
• What is energy infrastructure planning in cities and territories, and which plan-
ning steps and phases exist?
• What are the main requirements and quality factors of methods or methodolo-
gies supporting integrated planning in cities and territories?
• What is uncertainty in a model-based local planning context, and which types of
uncertainty exist?
• How are different types of uncertainty allocated according to planning and mod-
elling procedures?
• Which types of uncertainty are addressed in the planning literature, and which
methods and methodologies have been used to address different types of
uncertainties yet?
• Which method or set of methods based on which paradigm is most appropriate
for modelling and uncertainty analysis in a coherent, comprehensive and prag-
matic way?
• How does one avoid the strong cut consideration between a developed model
and its environment to avoid uncertainties and model complexity?
• What is the return of the experience for the proposed modelling and uncertainty
analysis approaches according to defined quality factors?

vii
viii Preface

This book has eight chapters. Chapters 1 and 2 introduce practical and research
contexts, problems, energy infrastructure planning and several background
­theories. Chapter 3 discusses a deterministic view of modelling and planning
proposing 3 domain-modelling concepts; Chaps. 4 and 5 discuss a stochastic view
of modelling and an uncertainty analysis suggesting two multi-method approaches
for uncertainty analysis based on probability and fuzzy set theories, respectively.
The applicability of the proposed multi-method approaches for modelling and
uncertainty analysis is illustrated in Chap. 6 in two different ‘Use Cases’ and is
then evaluated in Chap. 7. Chapter 8 concludes with the main outcomes, discusses
the limitations and proposes future work. Many different individual methods for
modelling and uncertainty analysis are reviewed and discussed according to the
defined quality factors in different chapters. Some additional empirical results are
presented in the appendixes.
In this volume, theories, methodological backgrounds and practical imple-
mentation in case studies will help not only understand the proposed methodolo-
gies for modelling and uncertainty analysis but also show how these approaches
are working in practice and the empirical outcomes and performance of these
approaches. The proposed approaches can be implemented not only in the context
of energy planning but also in other infrastructure planning, such as transportation
or water resources planning, and on the level of national planning.
We hope that business professionals, city or territory planners, researchers or
students using this book will have several advantages, e.g., learning, discovering
future research areas or implementing the proposed methodologies in real case
studies to cope with the complexity and uncertainty successfully.
This book has significantly benefited from the feedback, interactions and dis-
cussions in projects ‘City simulation platform development’ and ‘Smart and
low carbon cities’ of the European Institute for Energy Research (EIFER) and
Électricité de France (EDF) R&D. Thanks to all project participants and particu-
larly to Fabrice Casciani, Kevin McKone, Nico Schweiger, Susanne Schmidt and
Tobias Jäger for valuable suggestions and to Ludmila Gautier and Andreas Koch
for evaluation support.
Contents

1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Scope and Structure of the Book. . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Main Questions Addressed and the Purpose of the Book . . . . . . . . . 3
1.3 Overall Definitions and Theoretical Backgrounds. . . . . . . . . . . . . . . 5
1.3.1 Defining Planning, Scenarios, Strategies and Initiatives. . . . 5
1.3.2 Systems from the System Science Point of View. . . . . . . . . . 8
1.3.3 Models and Modelling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.4 Mixed Method Methodologies, a Pragmatic View. . . . . . . . . 12
1.3.5 Pre-existing Concepts of Uncertainty in Planning
and Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.6 Planning and Decision Making in Different
Information Availability Conditions. . . . . . . . . . . . . . . . . . . . 16
1.3.7 Theories for Uncertainty Analysis and Representation. . . . . 17
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 Energy Infrastructure Planning in Cities and Territories,

Quality Factors of Methods for Infrastructure Planning. . . . . . . . . . . 25
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Integrated Energy Planning in Cities and Territories. . . . . . . . . . . . . 26
2.3 Energy Systems in City and Territory,
a Sociotechnical Infrastructure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Defining Typology of Application or Use Cases. . . . . . . . . . . . . . . . 28
2.4.1 Use Case I: Decentralised Multi-model Based IEPCT . . . . . 28
2.4.2 Use Case II: Integrated-Model Based IEPCT . . . . . . . . . . . . 29
2.5 Modelling in IEPCT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5.1 Models and Different Degrees of Formalisation . . . . . . . . . . 29
2.6 Overall Requirements and Quality Factors of Energy
Planning and Modelling Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.7 Summary and Open Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

ix
x Contents

3 3-Domain Modelling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 3-Domain Metasystem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 3-Domain Modelling: Different Approaches
for Different Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.2 Data-Driven Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3.3 Process-Driven Modelling. . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3.4 Judgmental-Driven Modelling. . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Defining Modelling Approaches for Different Modelling Domains
and Use Cases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4.2 Modelling Approaches for Targeted Domain. . . . . . . . . . . . . 48
3.4.3 Data Driven Modelling Approaches for Neighbouring
and Distant Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.4 Modelling the Distant Domain and Its Impact
to Other Domains. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Summary of Modelling Approches for Different Use Cases
and Domains. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 3-Domain Modelling in Context of Multi Method Research. . . . . . . 63
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4 Conceptual Basis of Uncertainty in IEPCT . . . . . . . . . . . . . . . . . . . . . . 67

4.1 Why Be Explicit About Uncertainty in IEPCT? . . . . . . . . . . . . . . . . 67
4.2 Typology of Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2.1 Linguistic Uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.2 Epistemic Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.3 Variability Uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2.4 Decision Making Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2.5 Procedural Uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2.6 Levels of Uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Incorporating Uncertainty in Current IEPCT Studies . . . . . . . . . . . . 71
4.4 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5 Multi-method Approaches for Uncertainty Analysis. . . . . . . . . . . . . . . 73

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.1 IEP in Cities and Territories, Specific Conditions. . . . . . . . . 74
5.2 Analysis Sophistication Degrees . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.2.2 Appropriate Analytical Degrees in IEPCT Context. . . . . . . . 76
5.3 Quality Factors of Methods for Uncertainty Analysis. . . . . . . . . . . . 77
5.3.1 Technical Quality Factors. . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3.2 Organisational Capability. . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3.3 Satisfaction by Planning Participants. . . . . . . . . . . . . . . . . . . 78
Contents xi

5.4 Methods and Methodologies for Uncertainty Assessment:

A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.4.1 Evaluation Criteria. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.4.2 List of the Reviewed Methods and Methodologies. . . . . . . 80
5.4.3 Summary of Evaluation Results of Reviewed Methods. . . . 80
5.5 Multi Method Approaches for Uncertainty Analysis. . . . . . . . . . . . . 81
5.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.5.2 Fuzzy Scenario Based Uncertainty Analysis
for Use Case-I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.5.3 Probabilistic, Random Sampling Based Uncertainty
Analysis (PRSUA) Approach for Use Case-II. . . . . . . . . . . 90
5.6 A Review of Methods and Methodologies for Uncertainty
Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.6.1 Correlations and Copulas. . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.6.2 Expert Elicitation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.6.3 Fuzzy Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.6.4 Innovative Multimethod Approach (IMMA). . . . . . . . . . . . 104
5.6.5 Inverse Modelling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.6.6 Interval Prediction (IP) in Data Driven Models. . . . . . . . . . 107
5.6.7 Monte Carlo Simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.6.8 Multiple Model Simulation (MMS) of Process
Driven Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.6.9 Multiple Model Simulation (MMS) of Data
Driven Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.6.10 Scenario Analysis and Fuzzy Clustering . . . . . . . . . . . . . . . 115
5.6.11 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.6.12 Tests of Complex Models for Model Uncertainty. . . . . . . . 123
5.6.13 NUSAP and PRIMA Methodologies. . . . . . . . . . . . . . . . . . 125
5.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

6 Implementation of Discussed Uncertainty Analysis Approaches

in Case Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.1 Selection of Application Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.2 An Example of Use Case I: Singapore. . . . . . . . . . . . . . . . . . . . . . . . 132
6.2.1 Development of the “Singapore Sustainable
Growth” Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.2.2 Uncertainty Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.3 An Example of Use Case II: Mexico City. . . . . . . . . . . . . . . . . . . . . 152
6.3.1 Modelling Mexico City’s Waste-to-Energy System. . . . . . . 152
6.3.2 Uncertainty Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
xii Contents

7 Evaluation and Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

7.1 Evaluation and Discussion of the 3-Domain Modelling
Concept and Different Modelling Approaches. . . . . . . . . . . . . . . . . . 163
7.1.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
7.1.2 Modelling Approaches for Targeted Domain. . . . . . . . . . . . . 164
7.1.3 Modelling Approaches for Neighbouring Domain. . . . . . . . . 165
7.1.4 Modelling Approaches for Distant Domain. . . . . . . . . . . . . . 166
7.2 Evaluation and Discussion of Uncertainty Analysis Approaches . . . 166
7.2.1 General. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
7.2.2 Evaluation of FSUA Multi Method Approach
and Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.2.3 Evaluation of PRSUA Multi Method Approach
and Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.2.4 Comparative Assessment of Proposed Approaches. . . . . . . . 172
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

8 Overall Conclusion and Future Research. . . . . . . . . . . . . . . . . . . . . . . . 173

8.1 Overall Synthesis and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 173
8.2 Synthesis and Conclusions of Chaps. 1 and 2 . . . . . . . . . . . . . . . . . . 173
8.3 Synthesis and Conclusions of Chap. 3. . . . . . . . . . . . . . . . . . . . . . . . 174
8.4 Synthesis and Conclusion of Chap. 4. . . . . . . . . . . . . . . . . . . . . . . . . 175
8.5 Synthesis and Conclusions of Chaps. 5, 6 and 7 . . . . . . . . . . . . . . . . 175
8.6 Future Work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

Appendix A: Descriptive Analysis, Modelling of Historical Data. . . . . . . . 179

Appendix B: Some Empirical Results of Use Case I-Singapore. . . . . . . . . 183

Appendix C: Some Empirical Results of Use Case II-Mexico . . . . . . . . . . 193

Appendix D: C
 omparison Different Extrapolation,
Data Driven Methods and Intervals. . . . . . . . . . . . . . . . . . . . 199

Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
List of Figures

Figure 1.1 An example for positive membership function

for real number. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Figure 2.1 General procedure of IEPCT (adapted from [6],
used with permission of Elsevier). . . . . . . . . . . . . . . . . . . . . . . 27
Figure 2.2 IEPCT process and modelling steps
(adapted from [9] used with permission of Elsevier). . . . . . . . 30
Figure 2.3 Overall planning quality factors. . . . . . . . . . . . . . . . . . . . . . . . 31
Figure 3.1 Metasystem of modelling domains. . . . . . . . . . . . . . . . . . . . . . 41
Figure 3.2 Basic components of a stock-flow diagram . . . . . . . . . . . . . . . 49
Figure 3.3 Influence-dependence chart . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Figure 3.4 Allocation of descriptors and model stability. . . . . . . . . . . . . . 51
Figure 3.5 A typical FNN with one hidden layer. . . . . . . . . . . . . . . . . . . . 57
Figure 3.6 Some commonly used activation functions. . . . . . . . . . . . . . . . 58
Figure 3.7 Different impact types of causal forces
on the data-driven model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Figure 5.1 Themes and aspects influencing the design
of multi-method approaches for uncertainty analysis. . . . . . . . 74
Figure 5.2 FSUA approach for Use Case-I . . . . . . . . . . . . . . . . . . . . . . . . 83
Figure 5.3 PRSUA for Use Case-II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Figure 6.1 Qualitative “Singapore sustainable growth” model
and identification of key descriptors. . . . . . . . . . . . . . . . . . . . . 134
Figure 6.2 Evolution of descriptor importance ranking
according to influence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Figure 6.3 Population growth in Singapore multi model simulation,
composite extrapolation (Data, historical data; Theta, arima
(0.2.2), arima (1.2.2) is generated via auto.arima; nnetar,
neural networks; RandWalkFwD, random walk with drift
or ‘trend’; LinReg, Linear Regression, RobTrend,
Robust trend [see more details in Sect. 3.4.3].) . . . . . . . . . . . . 137

xiii
xiv List of Figures

Figure 6.4 Uncertainty of model driving forces, different intervals

(Normal, based on normality assumption of residue;
Chebychev, interval method based on Chebychev’s
inequality, see more details in Sect. 5.6.6). . . . . . . . . . . . . . . . 143
Figure 6.5 Consistency analysis of the targeted domain key
descriptors development of Singapore . . . . . . . . . . . . . . . . . . . 144
Figure 6.6 Cross-impact analysis of the targeted domain key
descriptors development of Singapore . . . . . . . . . . . . . . . . . . . 145
Figure 6.7 Trapezoidal membership functions implemented
in the study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Figure 6.8 Consistency diagram of different scenarios . . . . . . . . . . . . . . . 148
Figure 6.9 A priori and posterior probabilities of different
descriptors values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Figure 6.10 The number of clusters versus the objective function value. . . 150
Figure 6.11 Trade off among the partition coefficient and classification
entropy according to amount of cluster centres . . . . . . . . . . . . 151
Figure 6.12 The waste management system in Mexico city . . . . . . . . . . . . 153
Figure 6.13 System dynamics model of Mexico city’s
“Waste-to-energy” model [11] . . . . . . . . . . . . . . . . . . . . . . . . . 154
Figure 6.14 Tornado diagram of Mexico city’s waste-to-energy
model descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Figure 6.15 Population growth in Mexico, multi model simulation,
composite extrapolation (Data, historical data; Theta,
ARIMA (0.2.2), ARIMA (4.1.2) is generated via auto.arima;
NNAR(1), neural networks; RandWalkFwD, random walk
with drift; LinReg, Linear Regression, RobTrend, Robust
trend [see more details in Sect. 3.4.3]). . . . . . . . . . . . . . . . . . . 157
Figure 6.16 Syntax, and model components specification tests
of Mexico city’s waste-to-energy model. . . . . . . . . . . . . . . . . . 158
Figure 6.17 Description of uncertainty of some model inputs. . . . . . . . . . . 160
Figure 6.18 Model output uncertainty for the total waste generated. . . . . . 161
Figure A.1 Additive decomposition of min air temperature in Singapore. . . 180
Figure A.2 Histogram, distribution of population ages 65 and above
in Singapore . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Figure A.3 Single box plot, share of renewables in total energy,
in Mexico . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
List of Tables

Table 1.1 Characteristics of simple and complex system . . . . . . . . . . . . . . 9

Table 2.1 Planning tasks and time horizon in energy industry . . . . . . . . . . 26
Table 2.2 Requirements and quality factors of planning methods. . . . . . . . 32
Table 3.1 Preselected methods for extrapolation of driving forces. . . . . . . 52
Table 3.2 Reference impact matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Table 3.3 Relationship of causal forces to trends [70] . . . . . . . . . . . . . . . . 62
Table 3.4 Modelling methods, use cases and domains . . . . . . . . . . . . . . . . 63
Table 4.1 Uncertainty types in modelling and planning process ([1]
used with permission of ELSEVIER). . . . . . . . . . . . . . . . . . . . . 68
Table 5.1 Quality factors of the methods for uncertainty analysis . . . . . . . 79
Table 5.2 FCUA approach, planning and modelling steps . . . . . . . . . . . . . 89
Table 5.3 PRSUA approach and planning or modelling steps. . . . . . . . . . . 97
Table 5.4 An example of consistency matrix. . . . . . . . . . . . . . . . . . . . . . . . 116
Table 5.5 An example of cross-impact matrix. . . . . . . . . . . . . . . . . . . . . . . 117
Table 5.6 CIM and coefficients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Table 6.1 Final list of key descriptors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Table 6.2 Reference impact matrix for Singapore. . . . . . . . . . . . . . . . . . . . 139
Table 6.3 Average performance of data driven models and methods
across all data sets for some data of Singapore. . . . . . . . . . . . . . 142
Table 6.4 Linguistic terms for cross-impact value. . . . . . . . . . . . . . . . . . . . 146
Table 6.5 An example of structure of the Mamdani-style
inference implemented. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Table 6.6 Scenarios having high membership values to final
cluster centres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Table 6.7 Different scenarios for Singapore future development . . . . . . . . 151
Table 6.8 List of the 15 key descriptors of Mexico city waste
to energy model for some data of Mexico. . . . . . . . . . . . . . . . . . 156

xv
xvi List of Tables

Table 6.9 Average performance of data driven models and methods

across all data sets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Table 6.10 Description of uncertainty of some model inputs. . . . . . . . . . . . 160
Table 7.1 Comparative assessment of proposed and commonly
used uncertainty analysis approaches . . . . . . . . . . . . . . . . . . . . . 171
Abbreviations and Symbols

General Arithmetic, Logical Symbols, Relations and Operators


 Addition operator of real-valued variables
Multiplication operator of real-valued variables
∈ Element of
∀ Universal quantifier, for all
U, V Event
lim Limes, limit
W Sample space of the random events
x Absolute error
x̃ Estimated value of variable
εx Relative error of variable
x… Variables, x ∈ R
min(x) Minimum value of valuable
max(x) Maximum value of valuable
ϑ Truth value of variable
′
ϑ Normalized value of variable
x0 Initial model input value
y0 Initial model output value
∂y Model output change
∂x Model input change
UE (x, y) U uncertainty E elasticity of y output referred to model x input
fBt , fBn Behaviour function
Sx Sampling variable

xvii
xviii Abbreviations and Symbols

General Sets and Fuzzy Set Symbols, Neural Networks

⊆ Subset of
∩ Intersection
∪ Union
A, B Sets
max Maximum operator
min Minimum operator
µA (x) Membership function of A
Ac Complement of A
∅ Empty set
CoG Centre of gravity
c Amount of clusters
n Amount of objects
m Fuzzy exponent m ∈ (1,∞)
ui,j Degree of membership of object j to cluster i
d Distance between object and cluster centre i
xj Character vector of objects j, X = {�x1 , x�2
, . . . , x�n } 
βi Character vector of cluster centre i, B = β�1 , β�2 , . . . , β�c
PC(U) Partition coefficient of U cluster
PE(U) Partition entropy U cluster
ωch Weights for connections between constant input and hidden neurons
ωco Weights of the direct connection between constant input and output
ωih Weights between inputs and hidden neurons
ωho Weights between hidden neurons and output
fh , f0 Activation functions

Scenarios and Time Series Analysis

AIC Akaike’s information criterion

ANN Artificial neural networks
ARIMA Autoregressive integrated moving average
CIM Cross-impact matrix
MASE Mean absolute scaled error
MdAPE Median absolute percentage error
MICMAC Matrice d’Impacts Croiés—Multiplication Appliquée à un Classement
SSE Squares estimation procedure
et (n) Forecast error
β Intercept
ε Error
β0 , β1 Feting coefficients
θ Theta coefficient
Abbreviations and Symbols xix

σ2 Variance
PI Prediction interval
h Period
 n (h)
X Forecast of the series Xt
α Quintile of the distribution
Yi Observation at time t
Yi−1 Observation at time t − 1
n Time periods considered
K Key descriptor or factor
Sm m scenario
nj
kj Value of key factor kj
nj
c(knii · kj ) Consistency value of combination value knj
j
Csum (Sm ) Sum of consistency values of different combinations
PPrior A priory probability
Pnorm Posterior probability
Ni Number of potential adopters in the population
Ni Cumulative number of adopters
a Time shift parameter
k Constant of innovation diffusivity
b Diffusion rate parameter
p Order of autoregressive part (AR)
d Degree of first differencing involving
q Order of the moving average part (MA)
φ(B), θ(B) Polynomials
B Backshift operator
∝, β Smoothing parameters

Own and General Abbreviations

AB Agent-based simulation
AS Analytical sophistication
BA Boundary adequacy
Bel. Belief
C Correlations or copula
D Discontinuities
DC Dimensional consistency
DES Discrete event simulation
DDM Data-driven modelling
DM Decision Maker
EC Extreme conditions
EE Expert elicitation
FC Fuzzy clustering
FI Fuzzy inference
xx Abbreviations and Symbols

FL Fuzzy linguistic terms, scores

FM Fuzzy map
FSA Fuzzy-scenario-based assessment
FSUA Fuzzy-scenario-based uncertainty analysis
IEP Integrated energy planning
IEPCT Integrated energy planning in cities and territories
IM Inverse modelling
IMMA Innovative multi-method approach
IN-PA Inverse modelling
Int. Information intensity
IP Interval prediction
MCA Monte Carlo analysis
MMS Multi-model simulation
MSS Multi-software simulation
Nec. Necessity
NUSAP Numeral unit spread assessment pedigree
OSA One at time sensitivity analysis
PDF Probability density functions
PI Prediction interval
Pl Plausibility
Poss. Possibility
PRA Probabilistic risk assessment
Prob. Probability
PRIMA Pluralistic framework of integrated uncertainty management and risk
analysis
PRSUA Probabilistic, random sampling-based uncertainty analysis
PV Parameter verification
QA Quality assurance
RIM Reference impact matrix
SA Sensitivity analysis
SC Scenario analysis
SD System dynamics
SV Structure verification
TC Trends–cycles
TEST Model structure or behaviour tests
U Uncertainty
UA Uncertainty analysis
WdC Wild cards
WS Weak signals
Chapter 1
Introduction

The objectives of this chapter are

• Presentation of the main scope, objectives and hypothesis used in the book
• Presentation of an overview of the book with different main questions addressed
in the different chapters
• Providing unified definitions of different terms in the context of planning, mod-
elling, system and uncertainty analysis, and identifying and comparing different
concepts
• Presenting the principal concept of uncertainty in the public planning context
• Outlining the theoretical basis of implemented theories, such as fuzzy set the-
ory, probabilistic theory or mixed method research

1.1 Scope and Structure of the Book

The scope of this book is the modelling and uncertainty analysis of long-range
infrastructure development in cities and territories. The focus of the discussion is
concentrated on energy infrastructure as one of the critical parts of overall city and
territory infrastructure. Local energy infrastructure has numerous different sub-
systems with nonlinear, dynamic interactions. Usually, energy infrastructure plan-
ning is performed on the national level. Several methods and methodologies, such
as Integrated Resource Planning (IRP), Integrated Assessment of Supply, Demand-
Side Options (IASDO) and Least-Cost Planning (LCP), have been developed and
applied at the national level. The liberalization of energy markets in several coun-
tries, along with the growing emphasis on environmental protection and sustainable
development worldwide, has increased the interest in integrated energy infrastruc-
ture planning on the territory or city level [1]. The importance of using integrated
approaches for infrastructure planning in cities and territories has been recognized

© Springer International Publishing Switzerland 2015 1

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_1
2 1 Introduction

by the European Commission [2] and previous research [3–8]. Compared to other
planning approaches, which consist of either business planning (focused primar-
ily on profit) or policy planning (where public participation is limited), the specific
characteristics of Integrated Energy Planning in the Cities and Territories (IEPCT)
is that it is public planning involving different planning participants directly.
Additionally, and unlike policy planning, IEPCT takes place under constrained
budget conditions (e.g., the energy transition in Germany in 2011). IEPCT is a pro-
cess rather than a single task, involving different uncertainties, unclear problems
and local specificities. The main tendencies in IEPCT have been identified by [1]:
• “Within the cities and the territories, a growing community awareness of envi-
ronmental issues;
• Growing interest in the use of distributed generation technologies based on
renewable resources and small cogeneration systems;
• An increasing number of decision makers with different interests and prefer-
ences participating in the planning process;
• Development of a cross-sector analysis among several sectors such as industry,
households and transportation”.
Long-range planning and modelling of this type of infrastructure is a very complex
task containing many uncertainties. These tasks require methods and methodologies
that are progressively designed, presented and discussed in the following chapters of
this book. However, first and foremost, this chapter provides basic definitions that are
used along with the different chapters such as planning, scenarios, initiatives, systems
and uncertainty and introduces the overall considerations that support the discus-
sion regarding the methodologies, the design of which is presented in this book: the
system and its modelling; the motivations and framework for using a mixed meth-
odology approach for designing a uncertainty analysis methodology of long-range
infrastructure development in cities and territories; and the multiple concepts behind
uncertainty and its representation. Moreover, for the reader who would not follow a
linear reading of this book, the next section provides the main questions addressed
in each chapter. Based on a literature review, Chap. 2 highlights the general needs
and issues related to long-term infrastructure planning, particularly to IEPCT. These
needs and issues are then reformulated into general requirements and quality factors
of IEPCT methods. Once the requirements are stated (known), the design of methods
to support infrastructure planning can begin. The modelling of infrastructure devel-
opment in cities and territories raises many questions that are addressed in Chap. 3,
which introduces the 3-domain Metasystem framework and its implementation in
IEPCT: the 3-domain modelling. The 3-domain Metasystem framework allows iden-
tification of the appropriate modelling approach while taking into account simultane-
ous system consideration, multiple study perspectives and planning constraints such
as limited expertise or data. It is a framework for creating a dynamic and flexible
modelling environment. In addition, it allows a coherent, comprehensive and prag-
matic coverage of the planning problem. Based on the 3-domain Metasystem frame-
work, a 3-domain modelling concept is discussed. For different domains and ‘Use
Cases’ different modelling approaches are identified on the basis of the literature
1.1 Scope and Structure of the Book 3

discussions and experiences. However, the final evaluation of the modelling approach
for a given domain is discussed again in Chap. 7, after feedback from practical imple-
mentation is presented. To incorporate uncertainty analysis in IEPCT, it is essential to
clarify which types of uncertainties are present at the different stages of the IEPCT
process. By answering this question, Chap. 4 provides a framework for incorporat-
ing uncertainty types into an IEPCT methodology. Following the example of Chap. 2,
which provides a general requirement for IEPCT, Chap. 5 begins by defining the per-
formance criteria and quality factors of methods when intending to address uncer-
tainty categories in IEPCT. These definitions supply the design of a methodology
with additional constraints and requirements when aiming to incorporate uncertainty
analysis into the context of IEPCT. In addition, it allowed the review of available
methods for uncertainty assessment and choosing those which are used in the two
multi-method approaches for uncertainty analysis provided at the end of the Chap. 5.
The two proposed methodologies are devoted for two different IEPCT situations.
Their use is illustrated in two characteristic ‘Use Cases’ in Chaps. 6 and 7 discusses
the feedback we acquired of their use from the point of view of the different quality
factors. Finally, Chap. 8 summarizes the work and provides openings for future work.

1.2 Main Questions Addressed and the Purpose

of the Book

Questions related to overall groundings and organisation of the book in this

chapter
• What is the scope of the book and how is it organised?
• What are the definitions of planning, system, uncertainty, risk, scenarios, strate-
gies, cases and initiatives?
• What is a complex system? What is system thinking? How can we model com-
plex systems such as those concerned by IEPCT? Are there different types of
models? What are they?
• Should we design a monolithic approach or a mixed methods methodology? Is
there a generic framework for designing mixed methods methodologies?
• What are the pre-existing concepts of uncertainties in planning and modelling?
How is it linked to decision making? What are the main general theories for
uncertainty analysis and representation?
• What are the main questions addressed in the different chapters of the book.
Questions related to energy planning and modelling in cities and territories
in Chap. 2
• What is energy infrastructure planning in cities and territories? How is the inte-
grated energy planning modelling and planning process organised? What type
of planning and modelling steps and phases exist?
• What is energy infrastructure in cities and territories? Is it complex?
4 1 Introduction

• Which type of possible use cases exist?

• What are the main requirements and quality factors of methods or methodolo-
gies supporting energy infrastructure planning in cities and territories?
Questions related to modelling of complex energy infrastructure in cities
and territories in Chap. 3
• How is modelling of complex infrastructure development in cities and territories
achieved?
• How is low data availability dealt with in modelling while considering a deci-
sion makers interests?
• How are strong cut considerations avoided between the developed model and its
environment to reduce uncertainties and model complexity?
• What type of modelling approaches exist and how can these approaches contrib-
ute to the modelling of complex infrastructure in an appropriate way?
• Which approach can support a comprehensive, complementary and pragmatic
modelling of complex city or territory infrastructure planning?
Questions related to uncertainty concepts in the context of city and territory
infrastructure planning in Chap. 4
• Which typologies of uncertainty exist in IEPCT?
• How are different types of uncertainty allocated to planning and modelling
procedures?
• Which types of uncertainty are addressed in IEPCT, and which methods and
methodologies are used to address them in the IEPCT context?
• What are the open problems and questions related to uncertainty analysis in the
context of IEPCT?
Questions related to uncertainty analysis and development of uncertainty
analysis methodologies in Chap. 5
• What are the performance criteria and quality factors of methods for addressing
different uncertainty categories?
• How is it possible to design a methodological framework to address different
types of uncertainty in a systematic and holistic way?
• Which method or methodology is most appropriate for such a framework
according to the quality criteria and the type of Use Case?
Questions related to implementation of uncertainty analysis methodologies
in Chap. 6
• What is the performance of modelling methods implemented with 3-domain
modelling concepts?
• What is the return of the experience for the multi-method approaches for uncer-
tainty analysis?
1.2 Main Questions Addressed and the Purpose of the Book 5

Questions related to the final evaluation of uncertainty methodologies in Chap. 7

• What is the final performance of the uncertainty analysis methodologies and the
3-domain modelling concept according to the identified quality factors for anal-
ysis methods?
• Which modelling method in the 3-domain modelling concepts is most relevant
for the different planning conditions?
By addressing the above mentioned questions, the main purposes of the book
are to provide
an overall modelling approach for supporting complex infrastructure planning and
methodological frameworks for addressing different types of uncertainty in IEPCT
in a systematic and holistic way.
Different hypotheses are used and are discussed in the related chapters
or sections. The main hypotheses are:
• The availability of information and data for methods and methodologies to
support IEPCT will increase. However, it is assumed that long-range planning
remains in uncertainty conditions where data availability still will not be ade-
quate to precisely define future environmental states.
• Different types of uncertainties will remain in IEPCT in the future.
• The available computing power for implementation of methods and methodolo-
gies will increase.

1.3 Overall Definitions and Theoretical Backgrounds

One common way to deal with complex systems’ modelling like those involved in
IEPCT, is to use several approach from different domains. But each domain has its
own way of thinking, modelling implementing different terminologies. The same
word in one domain may have another meaning in another one, or the same word
may have multiple meanings. In order to clarify our working definitions this sec-
tion provides some useful definitions and theoretical backgrounds.
As told previously, this books aims at providing not only organized sets of
methods for modelling and for addressing uncertainties in IEPCT, but also keys
to design and evaluate them. Thus we found important to share with the reader the
overall underlying backgrounds of the design process and the choices we made.

1.3.1 Defining Planning, Scenarios, Strategies

and Initiatives

Planning, particularly strategic planning has different features. Mintzberg [9]

defines planning as (1) thinking about future, (2) controlling the future, (3) inte-
grated decision making. In general, planning is formulated as looking ahead, sys-
tematically thinking and formulating goals, developing alternatives for action, and
selecting and adopting them for efficient implementation [9].
6 1 Introduction

Planning can have different dimensions. According to the time horizon, it can
be short-, middle- or long-term planning [10]. According to the hierarchical aspect,
it can be strategic, tactical or operative planning [10]. In energy planning, the time
horizon and hierarchical aspects differ from usual business planning. These issues
are discussed in Sect. 2.2.
Laux [11] discussed the different planning dynamics: flexible, rigid and rolling
planning. In flexible planning, depending on the changes of the environmental con-
ditions, the plans are continuously fixed each time. In contrast to flexible planning,
in rigid planning, the plans are defined irrepealably at the beginning of the planning
period without pre-calculating optional plans in the case when the environmental
conditions change. An example of such planning is deterministic perfect foresight
planning. Rigid planning with plan revisions is called rolling planning. Plans are
revised over time and repeatedly adapting to the changing environmental condi-
tions. However, rolling planning is considered rigid planning because not all of the
changes in the later planning period are included and calculated at the beginning
of planning. Another specific planning category is robust planning, which can be
implemented in uncertain conditions. It is defined by different robustness criteria,
such as the robustness of information, the robustness of the result or the robust-
ness of the optimality. Interested readers can find more details about this category
in Scholl [10].
Depending on the planning domain, planning can be public, policy and
company (or business) planning [12]. The main feature of business planning is that
it is mainly cost or profit oriented, while public planning also considers different
aspects (e.g., social and environmental) and is performed not only according to
existing regulations (like business planning) but also according to public policies.
However, unlike policy planning, where planning is performed by administra-
tion and government, the public or affected parties can be involved directly in the
public planning process.
Planning can also take place on different levels (e.g., national, regional, urban
or neighbourhood levels).
Grünig et al. [13] considered strategic planning as a process. In community
planning, Williams [14] emphasized that “The new strategic discourse needs to
emphasize the process more than the content, the actors more than the structures.”
In management and company development, strategic planning is defined accord-
ingly as a systematic process for long-range strategy formulation and integrated
company development.
Planning can be described with different phases and sub steps. Mintzberg [9]
defines six overall planning steps:
1. Seeing the Need of Planning
2. Formulating Alternatives
3. Selecting the Best Alternative
4. Implementing the Best Alternative
5. Monitoring and Controlling Results
6. Pursuing versus Abandoning Plans and Planning
1.3 Overall Definitions and Theoretical Backgrounds 7

A strategy is defined by Mintzberg [9] as a plan, position or perspective. To main-

tain consistency with other definitions in this book and for relevance in energy
planning, we shall use the following working definitions:
A strategy is a policy or set of initiatives that can be implemented through time
to achieve desired objectives.
An initiative is a single action or intervention from a decision maker on the sys-
tem to improve it.
Combining several initiatives and different strategic logics is possible such as
the use of robust strategy or adaptive strategy.
Numerous definitions, schools and traditions exist for scenario development
and use [15]. Godet states [16] “The word ‘scenario’ is often misused and serves
to qualify any set of hypotheses.”
A scenario is a consistent, plausible description of how the system and/or its
environment may/could/should develop in the future. The set of hypotheses used
in the scenario definition must be consistent, plausible and different for different
scenarios.
Scenarios are not forecasts. As Godet [17] states, “Forecasting is the assess-
ment, with a degree of confidence (probability), of a trend over a given period”.
Scenarios do not forecast or predict the future with probabilities (precise proba-
bilities), but rather represent the different plausible and consistent futures. In this
book, the overall term foresight is used when discussing future studies in general
and for scenarios, forecasting or prediction.
Scenarios are not forecasts. As Godet [17] states, “Forecasting is the assess-
ment, with a degree of confidence (probability), of a trend over a given period”.
Scenarios do not forecast or predict the future with probabilities (precise proba-
bilities), but rather represent the different plausible and consistent futures. In this
book, the overall term foresight is used when discussing future studies in general
and for scenarios, forecasting or prediction.
A special case in territory planning is urban planning. Urban planning his-
tory is considered by Albeverio et al. [18] in three different periods. In the
late 19th century, planning was performed in centralized top-down way [19],
whereas in 1970, planning had a more participatory character. As Taylor [19]
states, “ […] local planning authorities were required to publicise applications
for planning permission, and in particular, consult with immediate neighbours
of such proposals”. In the last decades, the experiences shows that the growth
and behaviour characteristics in the city are more appropriate for explaining cer-
tain dynamics in cities than just homogenous land use structures. Some planning
activities are performed more and more in a bottom-up way and consider com-
plex phenomena like emergence or discontinuous change [18]. The reader inter-
ested in a detailed discussion about the history of ‘economic location theory’,
‘social physics’, and ‘geographical/spatial morphology’ in urban planning can
read [20] and [21].
The discussion in the book is focussed on energy infrastructure planning in rela-
tion with other territory and city systems such as waste. A more detailed discussion
regarding this is provided in Chap. 2.
8 1 Introduction

1.3.2 Systems from the System Science Point of View

One of the working hypotheses of this book is that we are dealing with complex
systems in the planning phase and that we will have to address different represen-
tations of it and consider several aspects that are provided below; therefore, this
section will provide some system theoretical discussion regarding systems and
complex systems.
Skyttner [22] states that the ‘scientific world view’ starts in the beginning of
the 18th century and that “Tradition and speculation were replaced by rational-
ism and empiricism with the assumption that natural phenomena can and must
be investigated and explained.” The moral world was separated from the scien-
tific world and what cannot be explained was a matter of ‘undiscovered’ science.
Reductionism becomes a pre-dominant doctrine, which argues that phenomena
on high level can be reduced to number of sub-basic elements and explained by
knowledge about these elements. The reductionism was inhered in the different
sciences such as physics, mathematics or social science. Many complex problems
such as traffic-system breakdowns, environmental disasters, accumulation of emis-
sions or exponential urbanization starting in the middle of the twentieth century,
that after approximately 200 years of success of classical science, the existing
explanation(s) do not provide satisfactory results. It was released among others
that the wholeness of the system cannot be explained by its parts any more [22].
This phenomenon has been previously recognised by Greek philosopher, Aristotle
(384–322 BC), who stated that, “The whole is more than the sum of its parts.”
Making an extensive review Skyttner [22] provides an overall definition of
General Systems Theory. From a basic science standard, deals on an abstract
level, with general properties of systems, regardless of physical form or domain
of application and are supported by its own metaphysics in Systems Philosophy.
As many theories, General system theory is based on underlying assumptions that
are also discussed by Skyttner [22]: “One of the basic assumptions embraces the
concept of order—an expression of man’s general need for imaging his world as
an ordered cosmos within an unordered chaos” [22].
Oxford Dictionary defines a system as “A set of things working together as
parts of a mechanism or an interconnecting network”. However, as stated in
Skyttner [22] “Any system exists in (or is unique to) the eye of the beholder (be
this a person or a group) and is associated with interests, […].” In the philosophi-
cal discussion, Cameron explains that the “System is defined as a mode of descrip-
tion, as a description of a conceptual holism”. An important property of holism is
function “Evidently, the function is ‘of’ the holism” [23]. A system can be material
or abstract. It is distinguishable from the environment for considerable length of
time.
Flood et al. [24] defines Systems thinking as “… a framework of thought that
helps us to address complex things in a holistic way. The formalization of (giving
an explicit, definite, and conventional form) this thinking is what we have termed
systems theory”.
1.3 Overall Definitions and Theoretical Backgrounds 9

The system can be defined as closed in a certain context if there is no relation

between the system and everything external to that system. Conversely, an open
system can have a relationship with its environment across a boundary in the form
of material, information or energy [24]. Identifying a system and setting its bound-
aries is usually not straightforward, but sometimes, such as in social and technical
systems, it is difficult to make a clear cut differentiation between the system and
its environment. The boundaries’ setting depends on the system properties but also
on the research or planning interests. Moreover, agreeing with [25], Flood et al.
[24] assert that defining the system involves starting from the interaction, such as
economical or physical ones, and not from the units.
Answering the question, “What is a systems’ science all about?”, Flood et al.
[24] state that “it is all about dealing with complexity.” Mitchell [26] agreed that
neither a single science of complexity nor a single complexity theory exists yet.
The author defines a complex system as “[…] a system in which large networks of
components with no central control and simple rules of operation give rise to com-
plex collective behaviour, sophisticated information processing, and adaptation via
learning or evolution.” In an attempt to develop a transportable conceptual frame-
work of complexity, Flood [27] shows that complexity is related not only to sys-
tems via the number of nonlinear, asymmetric relationships of elements but also to
people via their notions, perceptions, interests and capabilities.
Based on this review, Skyttner [22] emphasizes the main characteristic differ-
ences between simple and complex systems Table 1.1. He also states that a sim-
ple system can also have complex unexpected behaviour in certain situations and
times.
By observing a city, infrastructure is a system possessing almost all of the char-
acteristics of complex systems. Moreover, as it is a socio-technical system, the
definition of its boundaries is not always easy and straight. Thus, it was neces-
sary to think about tools for modelling the system and subsystems of a city infra-
structure for IEPCT. Section 1.3.3 will provide the fundamentals of modelling and

Table 1.1  Characteristics of simple and complex system

Simple systems are characterized by Complex systems are characterized by
A small number of elements A large number of elements
Few interactions between the elements Many interactions between the elements
Attributes of the elements are predetermined Attributes of the elements are not
predetermined
Interaction between elements is highly Interaction between elements is loosely
organized organized
Well-defined laws govern behaviour They are probabilistic in their behaviour
Subsystems do not pursue their own goals Subsystems are purposeful and generate their
own goals
The system is unaffected by behavioural The system is subject to behavioural influences
influences
The system is largely closed to the The system is largely open to the environment
environment
10 1 Introduction

terminologies. This discussion is used in Chap. 3 to develop a 3-domain modelling

concept, which is based on the notion of 3-domain Metasystem consideration. The
3-domain modelling notion presents an integration of different modelling methods
used in different domains to model the system in an inclusive, coherent and prag-
matic way for a given planning context.

1.3.3 Models and Modelling

Models and modelling play an important role in infrastructure planning. They are
implemented in different branches such as energy system models, transportation
models for prediction, and decision support or analysis of complex behaviour. The
definitions and classification of models discussed in this section will support the
design and identification of different modelling approaches for the proposed 3
domain Metasystem framework in Chap. 3.
Numerous definitions of the term models and modelling exist in different disci-
plines. A very general definition of a model is provided in Zahn [28]. Accordingly,
a model is defined as a representation of something for someone and for a par-
ticular purpose. In the context of environmental modelling and in energy panning,
the modelling process is presented using different model building steps and dif-
ferent model definitions according to the modelling stage. Mirakyan and De Guio
[29] present the modelling process starting from mental models to applied models.
Refsgaard et al. [30] also use different model terms according to the modelling
stage such as conceptual model, model code or model.
In the planning context, models are a simplified representation of the real sys-
tem (e.g., energy system) relevant to give planning context.
Stachowaik [31] defines three main features of a model: representation, abbre-
viation, and pragmatics.
Representation feature
If all attributes of considered system elements and the planning system as a whole,
is similar to the model and vice versa then the model is isomorphic [31]. However,
in planning, a model is an abstraction of the real system and only relevant ele-
ments, attributes and relations of system are considered. These types of models are
homomorphic [31].
Abbreviation feature
The abbreviation might be performed in two steps: abstraction and relaxation. In
the abstraction abbreviation, only parts of the system that are relevant within the
planning context are considered. In the relaxation abbreviation, complex interrela-
tions in the model are relaxed, e.g., making dynamic models static or a nonlinear
interrelation linear [32], which makes the models more operational. Models can
have also features that are not available in the original object. These type of mod-
els are called abundant [31].
1.3 Overall Definitions and Theoretical Backgrounds 11

Pragmatic feature
As defined before, a model is a representation of something (a planning sys-
tem) for someone and is utilised for a particular planning purpose and under cer-
tain conditions. The planning conditions, such as the availability of data, time,
resources or expertise of a planner or modeller, can vary significantly. The goal of
modelling is to have a flexible and manageable model for certain planning condi-
tions. Elert et al. [33] states that “The complexity of a model must be weighed
against the availability and uncertainty of the input data because the apparent
advantage of using a more complex model, which may give a better representation
of reality, may be overshadowed by the introduction of greater data uncertainty.”
Modelling approaches can be classified in many ways. According to the pur-
pose (goal) of modelling in the context of planning, models can have different
determinations [12]. On the one hand, for the analysis of a historical or current
situation, modelling can support the development of models of a real system or
planning object, e.g., analysing the behaviour of current or historical model behav-
iour. On the other hand, it might be necessary to analyse the future states of a real
system. In this case, foresight-models are required. For the impact assessment or
appraisal, assessment-models are required to implement certain value systems,
such as the life cycle assessment model. Another category of models discussed by
Zürni [12] are decision-models. These models provide direct recommendations to
decision makers to address certain situations.
Modelling approaches can also be classified from a structural and behavioural
[24] point of view. In the structural approach, system elements are defined a pri-
ori. In the behavioural approach, only a particular type of interaction of interest is
chosen and this is then used to identify the structure of the system.
Zürni [12] provides a matrix of nine model types depending on the input cate-
gory and the interconnections of the model variables. Examples of these categories
are ‘semi-quantitative system models’ or ‘quantitative simulation models’. If the
interdependency among the model variables are defined via the ordinal scale (e.g.,
no impact, low or high impact), the models are defined as system models. If the
relations among the variables of the model can be presented using equations, the
models are called simulation models.
Depending on the mathematical formulation, modelling methods can be, e.g.,
linear programming or nonlinear programming.
According to modelling perceptive, modelling can be performed via top-down
or bottom-up. In bottom-up modelling, sometimes also referred to as ‘engineering-
type’ models, the description of a given system’s elements and their behaviour are
typically aggregated and represented for the whole system. Most of the bottom-
up models are highly disaggregated. In contrast, the top-down modelling approach
starts with the macro-economic performance of the economy as a whole, and then
the behaviour of subsystem elements are disaggregated and simulated.
Extensive reviews of different modelling methods implemented in city or terri-
tory planning are provided by [1, 34, 35].
12 1 Introduction

Modelling in territories, particularly the urban modelling in the 1960s and

earlier was aggregative, cross-sectional, non-behavioural and did not provide
dynamic view of cities. Those modelling approaches represented a fusion of social
physics ideas with rudimentary regional economics as developed within regional
science. Later in the 1980s and 1990s under the inspiration of complexity theory
urban models based on cells and agent began to appear [20]. However, Batty [20]
states that “As one moves more towards practical applications, there is less con-
cern for new theoretical developments but as urban models are still so idiosyn-
cratic in their design and construction”. The simulation now is not anymore to
predict single future but many futures because of intrinsically complex and inher-
ently unpredictable nature of urban development.

1.3.4 Mixed Method Methodologies, a Pragmatic View

1.3.4.1 Introduction

As we are dealing with complex systems, the questions whether we should build a
single method, how to use the pre-exist methods developed for uncertainty analy-
sis in different domains, how to integrate them in a consistent way for given goals,
how to prepare its evolution appeared early. We decided to take a multi methodol-
ogy approach for many reasons that are provided bellow.
Intensive research and discussions about the mixing of methods started in the
1960s. Abbas et al. [36] presented the limitations of using one method exclusively
within any of the predominant paradigms, qualitative (constructivist) or quantitative
(positivist or naturalist), and argued for using multiple methods (pragmatic para-
digm) to study complex problems. The same authors [37] define mixed-methods as
follows: “Mixed method studies are those that combine the qualitative and quanti-
tative approaches into the research methodology of a single study or multi phased
study”. Mingers [38] defined mixed methods as follows: “Multimethodology just
means employing more than one method or methodology in tackling some real-
world problem”. The key feature of mixed-method research is methodological plu-
ralism. The theoretical foundations of mixed-methods research, which has been
discussed, among others in [38–40], state three favourable situations for using
multi-methodology: first, in an intervention where different aspects, such as techni-
cal, ecological or social aspects, have to be taken into account; second, an interven-
tion that is not a single event but a process with different phases; and third, when
triangulation of different methods can increase the validity of results and provide
more confidence. All these conditions are present in IEPCT.
Five rationals for using mixed methods have been identified in [41]: triangula-
tion, complementarity development, initiation and expansion.
Triangulation seeks convergence, corroboration and correspondence of results
from different methods investigating the same phenomena [41]. The assumption
behind triangulation is that implementing heterogeneous methods for the same
problem will increase the validity of the results.
1.3 Overall Definitions and Theoretical Backgrounds 13

“Complementarity seeks elaboration, enhancement or illustration and clarifica-

tion of the results from one method with the results from the other method” [41].
Complementarities increase the interpretability and meaningfulness of examining
different overlapping facets of a phenomenon.
“Development seeks to use the results from one method to help developing or
informing the other method” [41]. Methods, qualitative or quantitative, can be
implemented sequentially or iteratively, wherein results of a qualitative analysis
can be implemented in quantitative modelling and vice versa.
“Initiation seeks the discovery of paradox and contradiction, new perspectives
of frameworks, the recasting of questions or results from one method with ques-
tions or results from the other method” [41].
“Expansion seeks to extend the breadth and range of inquiry by using different
methods for different inquiry components” [41]. Expansion increases the quality
of the investigation by selecting the most suitable method for different tasks.
In strategic management, mixed methods have been mainly implemented for
triangulation, complementarity, development and expansion yet [42].
There are several advantages to using different methods in combination instead
of using a single method. Molina-Azorίn [42] indicates several advantages of
using a mixed methods approach in strategic management:
• Provides better understanding of complex problems,
• Increases confidence in the validity of results,
• Depending on resource, time or data availability, different methods can be used
at different stages to support evolutionary process,
• Likely to be context and process-oriented,
• Presents divergent views,
• Provides a complete picture of the phenomena,
• Meets the needs of multiple audiences.
However there are also some drawbacks to using mixed methods [42]
• It may require extensive time, resources, and efforts,
• Researchers or planning participants may need to develop a broader set of skills.
The discussed uncertainty analysis approaches in Chap. 5 based on the experience
and theoretical foundation of mixed-method research.

1.3.4.2 Aspects for Designing Mixed Methods

Mingers [39] provides the activity associated with social, personal and material
dimensions of a project for each major stage of an intervention. The different
phases of intervention that have been proposed are:
• “Appreciation of the problem situation as experienced by the agents involved.
• Analysis of the underlying structure/constraints generating the situation as
experienced.
14 1 Introduction

• Assessment of the ways in which the situation could be other than it is; of the
extent to which the constraints could be altered.
• Action to bring about desirable changes”.
These stages are not discrete, sequential steps but “they are aspects of the interven-
tion that need to be considered throughout, although their importance will differ as
the project progresses”. These stages fairly correspond to IEPCT phases discussed
in Mirakyan and De Guio [1].
Multidimensionality is presented in Rosenhead and Mingers [40] in three
dimensions: the material world which is outside of human beings; the personal
world consisting of individual thoughts, emotions, experiences and beliefs; and the
social world, where individuals are members and share a certain social system.
Creswell [43] suggests four different aspects which influence the design of
a mixed-method study (timing, weighting, mixing, Theorizing\Transforming
Perspectives). These aspects are discussed below.
Timing indicates whether qualitative or quantitative methods should be used
sequentially or simultaneously, and the intervention phase in which they are com-
bined. The choice of the first method to be used depends on the problem context
and the intent of the participants [43]. Collecting participants’ qualitative judg-
ment at the beginning of the study leads to a study design that focuses on the
participants’ view to understand the phenomena. On the other hand, a deductive,
quantitative approach collects quantitative data to confirm certain hypotheses or
research questions. A sequential procedure can be repeated continuously, making
the process iterative e.g., qualitative → quantitative → qualitative. Qualitative and
quantitative approaches can also be implemented simultaneously or concurrently,
which might be a more practical approach in some cases for results validation.
Weighting refers to the priority assigned to qualitative or quantitative methods
or data. Weighting could either be equal for each metric or higher for one metric
compared with others [43]. The priority can depend on planning participants’ pref-
erences but also on the quality of information and data source available.
Mixing indicates the integration of qualitative and quantitative methods, the
phases in which they are mixed and their usage in combination. The different
phases include problem framing or questions definition, information collection,
analysis and assessment or interpretation and presentation of results [43].
Mingers [39] distinguishes different ways of mixing methods:
• Whether only quantitative or only qualitative methods or both are mixed,
• Whether certain elements of different methods or methodology are mixed,
which is called Multi-methodology [39] or else methods are mixed as a whole
which is called methodological combination [39]
• Whether one method supports or enhances other primary methods which is
called Methodology enhancement [39]
Additionally, mixing can also be differentiated by the manner in which methods
are linked
1.3 Overall Definitions and Theoretical Backgrounds 15

• Whether methods are linked hard or loose. In the hard link case, methods are
dependent on each other, have feedback and are linked in a dynamic way. In the
loose link case, there is no dynamic link and no information feedback.
Theorizing or Transforming Perspectives aspect indicates whether certain theo-
retical perspective guides the entire mixed-method design [43]. Theories might be
explicitly or implicitly mentioned. Theories shape the types of questions asked, the
participants in the study, the mode of data collection, and the interpretations made
from the study [43].
Some of multi-methodology designing aspects and rationales will be used in
Chap. 5 when designing multi-method approaches for uncertainty analysis.
As it will be seen in Sect. 1.3.4 we decided to use a multi-method approach for
analysing uncertainty in IEPCT. In our context we will show that the multi-method
methodology that has to be designed consists of a set of pre-existing and new
methods that have to work together in order to provide global and robust results
and to perform expected requirements.

1.3.5 Pre-existing Concepts of Uncertainty in Planning

and Modelling

One of the main discussion point in the book is the uncertainty analysis in context
of IEPCT, therefore the basic notion and definitions of uncertainty is discussed in
this section based on literature review and our experiences.
In the integrated assessment and environmental literature several taxono-
mies and concepts of uncertainty have been proposed and numerous definitions
of uncertainty have been provided in recent years [44]. However, there is still no
common agreement on typology and sources of uncertainties.
Van Asselt and Rotmans [45] proposed a concept for the sources of uncertainty
in integrated assessment. They classified uncertainty into two main sources that
are further divided into subcategories. These two sources of uncertainties have
been defined as: (i) uncertainty due to variability (inherent randomness of nature,
value diversity, behavioural variability, societal randomness, technological sur-
prise). Increased knowledge may narrow the variability, e.g., uncertainty interval,
but it cannot reduce it completely; and (ii) uncertainty due to limited knowledge
(inexactness, lack of observations/measurements, practically immeasurable, con-
flicting evidence, indeterminacy).
Walker [46] provided a generic, detailed conceptual framework for defining
uncertainty and categorisation for model-based decision support. Similar categories
of uncertainties have been discussed in Refsgaard [47]. Walker [46] also provided
a broad definition of uncertainty as being any departure from the unachievable
ideal of complete deterministic knowledge of the system. Norton [48] submitted
that “Modellers’ view of uncertainty is considered rather than a decision makers’
perspective” and “classification scheme omits some relevant sources of uncertainty
16 1 Introduction

(perhaps even the most important ones) that arise before and after scientific mod-
els are applied”. Additionally, “Insufficiently definition the ‘level’ dimension of
uncertainty like statistical dimension since distinct spectrum of well-established
methods, not all statistical, for characterising degrees of credibility, ranging from
bounds (binary classification as possible/impossible) through rough sets (ternary
classification as possible/doubtful/impossible), fuzzy sets.”
Ascough et al. [44] presented a review of the status of uncertainty analysis in envi-
ronmental modelling and suggested a new framework for uncertainty typology that
denote certain critical challenges. To some extent, their framework, which could be
considered an extension of the concept of Walker et al. [46], includes uncertainties in
not only related to model building but to the whole model-based decision-making pro-
cess. Ascough et al. propose an additional uncertainty categories like decision mak-
ing or linguistic uncertainty. However, their framework does not contain the levels of
uncertainty, such as statistical, scenario and ignorance, as suggested by Walker [46].
In public decision-making, Klauer and Brown [49] consider uncertainty as a
subjective property distinguishing different levels of imperfect knowledge in pub-
lic decision-making: Ignorance, Uncertainty, Error and Risk situations.
Ignorance is accordingly defined as follows [49]: “A person is ignorant with
respect to an event if they are unaware of the (potential) outcomes of that event or
of the event itself.”
Uncertainty is defined as follows [49]: “A person is uncertain if they lack confi-
dence about the specific outcomes of an event. Reasons for this lack of confidence
might include a judgement of the information as incomplete, blurred, inaccurate or
potentially false.” However, it is also indicated that uncertainty is not only a sub-
jective property but mainly caused by objective reasons.
Risk or risk situation is defined as follows [49]: “A risk situation is a person’s
representation of an event, where they assume to know all potential outcomes as
well as the probabilities of each outcome.” Most common approaches to address
risk are probability-based methods.
Error is defined as follows [49]: “Assume an event can be quantified within the
representation of the subject by
a value x̃ of the variable from the true value x. There are two types of errors:
the absolute error, ∆x =x̃ − x and
the relative error, εx x = x̃ − x /x”.
Error concept is very common in ex post analysis [49].
These definitions will be used in this book as fundamentals for the design of
uncertainty analysis frameworks.

1.3.6 Planning and Decision Making in Different

Information Availability Conditions

According to the classical decision theory three conditions (information availabil-

ity) are possible where decision making and planning is performed:
1.3 Overall Definitions and Theoretical Backgrounds 17

1. Decision in certain environment: in this case the situation, which environmen-

tal conditions will occur is clear (well known) ‘a priori’
2. Decision in risk environment: in this case the true state is not known but sub-
jective or objective probabilities for the occurrence of the various environmen-
tal conditions are known
3. Decision in uncertain environment: No probability information about the states
of environment is available in this case. It is only known that some of the envi-
ronmental conditions might occur.
Rommelfanger [50] identifies additional conditions reviewing recent literature.
The planning and decision making can accordingly take place in
4. Risk environment with fuzzy probabilities: in this case the probability about
states of environment can be only roughly stated. This case includes also situa-
tion when linear partially information is only available
5. Risk environment with possibilities: The possibility values about states of envi-
ronment are known
6. Risk environment with belief and plausibility function according to evidence
theory.

1.3.7 Theories for Uncertainty Analysis and Representation

Uncertainty analysis and representation have been traditionally performed using

probability theory. However, different theories exist for modelling and depicting
uncertainty. Halpern [51] presents and compares different uncertainty analysis
theories, such as possibility, evidence or probability theory. In this section, a brief
discourse on the basic notions of some theory will be provided.

1.3.7.1 Basic Notions of Probability Theory

Different theories of probability are discussed in Weatherford [52]. Except the

classical theory of probability, other probability theories, such as frequency, sub-
jectivist or Bayesian theories are discussed. An essential property of a probability
function is the additivity of the probabilities of mutually exclusive events.
Classical and frequentist theory of probability
The classical theory of probability of an event U, Prob(U), is usually implemented
for a finite number of cases which are equally likely to appear:
Number of cases of interests U
Prob(U) =
Total number of cases
Classical theory of probability is founded on Laplace’s principle of insufficient
reason—“knowing nothing at all about the true state of nature” is equivalent to
18 1 Introduction

“all states having equal probability” [53]. It is used in everyday life as an intuition
of what is likely to happen.
Probability is defined in relative frequency theory as the fraction of cases in
which the event U appears nU if the situation under consideration was repeated an
infinite number of times n.
nU
Prob(U) = lim
n→∞ n

Subjectivist or Bayesian theory of probability

The subjectivist or Bayesian view of probability of an event is the degree of
belief that a person has based on the relevant information available. The judgment
depends not only on a personal view of the event U but also on the available infor-
mation i. Thus, it is a function of two arguments
Prob( U|i)
Different people might assign different probabilities to the same event at different
times. Therefore, subjective probability is not an objective measure. Furthermore,
subjective probability assignment must be consistent with the same axioms of
classical probability. E.g., if U is assigned to have Prob(U) probability then the
complement U c event must be assigned 1-Prob(U) probability.

1.3.7.2 Basic Notions of Fuzzy Set and Possibility Theory

In classical set theory, an element can either belong to a set or not. Fuzzy Set-
Theory is a generalization of the classical set theory. Belohlavek et al. [54] states
that “ […] generalizing classical set theory to fuzzy set theory was to introduce
the capability for dealing with categories that lack sharp boundaries in a rigorous,
mathematical way”.
Fuzzy set theory was first published and mathematically formulated by Zadeh
in 1965 [55, 56]. He introduces new concept for applying and manipulating natu-
ral language in fuzzy terms. Several authors like [57–60] works on development of
Fuzzy set theory in different application.
In fuzzy theory, fuzzy set A of universe W is defined by function µA (x) called
the membership function of set A. The question is now not whether an element x
belongs to set A or not like in crisp set, but to which degree element x belongs to
set A. Mathematical it can be defined as follow
µA (x) : X → [0, 1],
Where
µA (x) = 1 if x is totally in A
µA (x) = 0 if x is not in A
1.3 Overall Definitions and Theoretical Backgrounds 19

0 < µA (x) < 1 if x is partly in A

Membership function µA (x) equals the degree to which x is an element of set A.
The degree, a value between 0 and 1, represent the degree of membership, which
is called membership value, of element x in set A. An example of membership
function for real number is presented in Fig. 1.1.
There are different types of membership functions. Most common used func-
tions are trapezoidal, singleton, Gaussian or triangular.
Different application has been developed based on Fuzzy set theory like Fuzzy
inference approach or fuzzy clustering. These approaches are implemented and
discussed in Chaps. 5 and 6.
Possibility theory was developed as a branch of Fuzzy set theory [61].
Possibility theory is reported to be a natural approach for modelling and repre-
sentation of non-precise information. A previous review [62] indicates that, among
the various interpretations, the most common and useful interpretation of possi-
bility theory was performed in terms of interpretation of fuzzy sets, which was
suggested by Zadeh. Zadeh interprets membership functions of fuzzy sets as possi-
bility distributions encoding elastic constraints induced by natural language state-
ments [63].
A possibility measure ‘Poss.’ associates with each subset of ‘W ’ a number in [0,
1] and satisfies the following three properties [51]:
(I) Poss(∅) = 0
(II) Poss(W ) = 1
(III) Poss(U ∪ V ) = max(Poss(U), Poss(V )) if U and V are desjoint.

Fig. 1.1  An example for positive membership function for real number
20 1 Introduction

Whereas the probabilities of an event and its complement must add to 1

(law of excluded-middle) Prob(U) + Prob Ū = 1∀U ∈ W , possibility of
an event and its complement add to a number greater or equal to unity
Poss(U) + Poss Ū ≥ 1∀U ∈ W [64]. The complement of possibility of con-
trary event is interpreted as a degree of necessity of the event. This means that
possibility of U does not inform us fully about the impossibly of its complement.
Necessity function is defined therefore to present the belief in the impossibility of
U c. Ness(U) = 1 − Poss(U c ) and Ness(U) ≤ 1 − Ness(U c ).
Presented in layman terms, possibility is a softer measure than probability.
What is probable must be possible; the opposite is not always valid. What is not
possible is also not probable [65].
In contrast to probability theory, which represents the probability of a certain
future event, possibility theory represents the degree to which the occupancy of
some event is possible.
Many authors, particularly Dubois and Prade [66, 67] work on the development
of possibility theory.

1.3.7.3 Basic Notion of Evidence Theory

Dempster-Shafer theory of evidence, originally presented by Arthur Dempster [68]

derived and referred to as the upper and lower probabilities of a given event. Glenn
Shafer [69] reinterprets the upper probability as plausibility and lower probability
as the epistemic degree of belief in the truth of an event. Shafer’s view, “belief” in
a hypothesis does not measure the chance that it is true, but rather the strength of
the arguments we have in favour of the hypothesis [70]. It is a generalization of
the subjective probability theory for each question of interest. However, evidence
theory involves two specifications, a belief and plausibility.
Degree of evidential support of subset A of W universe is represented by m(A),
which satisfies the following axioms [71]

0 ≤ m(A) ≤ 1, m(∅) = 0, {m(A)|A } = 1

m(A) represents a portion of total belief mass assigned (BMA) to A on the base of
available piece of evidence, which represents how strongly A is supported by evi-
dence. Total belief committed Bel(A) to A and plausibility Pl(A) that one does not
commit to the negation of A are defined by
 
Bel(A) = m(B), Pl(A) = m(B)
B⊆A B∩A� =φ

Future the belief Bel(A) in the occurrence of an event and the plausibility Pl(Ac ) of
the non-occurrence of an event must sum to one [72]
 
Bel(A) + Pl Ac = 1
1.3 Overall Definitions and Theoretical Backgrounds 21

Presented in layman terms, a piece of pure evidence proves a statement with a cer-
tain probability but has no opinions on its negation, therefore
 
Bel(A) + Bel Ac ≤ 1
In contrast to probability theory where occurrence and non-occurrence of an event
must sum to one, specification of plausibility is capable of incorporating recogni-
tion of alternatives that might manifest in the sum of the plausibility in the occur-
rence Pl(A) and non-occurrence Pl(Ac ) of an event being greater than one [72].
 
Pl(A) + Pl Ac ≥ 1
Despite the advantages having many intuitively properties, some shortcomings of
implementation of evidence theory have been discussed in the literature. Pearl [70]
concludes that “The BF (belief function) formalism encounters difficulties repre-
senting domain knowledge, […]”. He founds that “[…] belief functions have dif-
ficulties representing incomplete knowledge, primarily knowledge expressed in
conditional sentences (when evidences are in conflict)”. In a book review Zadeh
[73] states “In particular, the theory does not address the issue of chaining, nor
does it come to grips with the fuzziness of probabilities and certainty factors.”

References

1. Mirakyan A, De Guio R (2013) Integrated energy planning in cities and territories: a review of
methods and tools. Renew Sustain Energy Rev 22:289–297. doi:10.1016/j.rser.2013.01.033
2. Towards a thematic strategy on the urban environment ((COM(2004)60)) (2004) European
Commission. http://ec.europa.eu/environment/urban/towards_com.htm#language_versions
3. Nicholas Lenssen (1996) Local integrated resource planning: a new tool for a competitive
era. The Electr J 9(6):26–36
4. Bhatt V, Friley P, Lee J (2010) Integrated energy and environmental systems analy-
sis methodology for achieving low carbon cities. J Renew Sustain Energy 2(3).
doi:10.1063/1.3456367
5. Nilsson JS, Mårtensson A (2003) Municipal energy-planning and development of local
energy-systems. Appl Energy 76(1–3):179–187
6. Sperling K, Hvelplund F, Mathiesen BV (2011) Centralisation and decentralisation in strate-
gic municipal energy planning in Denmark. Energy Policy 39(3):1338–1351
7. Poupeau F-M (2010) Central-local relations in french energy policy making and the environmen-
tal shift: towards a new pattern of governance between state and local authorities? Sciences Po
8. Ravetz J (2000) Integrated assessment for sustainability appraisal in cities and regions.
Environ Impact Assess Rev 20(1):31–64. doi:10.1016/s0195-9255(99)00037-2
9. Mintzberg H (2000) The rise and fall of strategic planning. Financial Times Prentice Hall,
New Jersey
10. Scholl A (2001) Robuste Planung und Optimierung: Grundlagen - Konzepte und Methoden -
Experimentelle Untersuchungen. Physica-Verlag HD
11. Laux H (2005) Entscheidungstheorie. Springer, Berlin
12. Zürni SU (2003) Möglichkeiten und Grenzen der Methode der Szenarioanalyse in der
Öffentlichen Planung: eine Analyse am Beispiel der Schweizer Energieplanung wiku-verlag
13. Grünig R, Clark A, Kühn R (2010) Process-based strategic planning. Springer, Berlin
14. Williams PM (2002) Community strategies: mainstreaming sustainable development and
strategic planning? Sustain Develop 10(4):197–205. doi:10.1002/sd.197
22 1 Introduction

15. Durance P, Godet M (2010) Scenario building: uses and abuses. Technol Forecast Soc Chang
77(9):1488–1492. doi:10.1016/j.techfore.2010.06.007
16. Godet M, Roubelat F (1996) Creating the future: the use and misuse of scenarios. Long
Range Plann 29(2):164–171. doi:10.1016/0024-6301(96)00004-0
17. Godet M, Coates JF (1994) From anticipation to action: a handbook of strategic prospective.
Unesco Publishing, Paris
18. Albeverio S , Andrey D , Giordano P, Vancheri A (2008) The dynamics of complex urban
systems. Physica-Verlag Heidelberg and Accademia di Architettura, Mendrisio, Switzerland,
Heidelberg doi:10.1007/978-3-7908-1937-3
19. Taylor N (1998) Urban planning theory since 1945. SAGE Publications, Thousand Oaks
20. Albeverio S , Andrey D , Giordano P, Vancheri A (2008) The time-lines: cities, planning,
modeling. In: The dynamics of complex urban systems. Physica-Verlag Heidelberg and
Accademia di Architettura, Mendrisio, Switzerland. doi:10.1007/978-3-7908-1937-3
21. Isard W (1956) Location and space-economy: a general theory relating to industrial location,
market areas, land use, trade, and urban structure. Published jointly by the Technology Press
of Massachusetts Institute of Technology and Wiley, New York
22. Skyttner L (2005) General systems theory: problems, perspectives, practice. World Scientific,
Singapore
23. Cameron B (2013) Systems for philosophy. In: Ellis K, Gregory A, Mears-Young BR,
Ragsdell G (eds) Critical issues in systems theory and practice, p 712
24. Flood RL, Carson ER (1993) Dealing with complexity: an introduction to the theory and
application of systems science. Springer, Berlin
25. Reynolds PA (1980) An introduction to international relations. Longman, Harlow
26. Mitchell M (2009) Complexity: a guided tour. OUP, USA
27. Flood RL (1987) Complexity: a definition by construction of a conceptual framework. Syst
Res 4(3):177–185. doi:10.1002/sres.3850040304
28. Zahn E (1991) Strategieunterstützngssysteme, In: Milling P (ed) Systemmanagement und
Managementsysteme. Duncker & Humblot, Berlin
29. Mirakyan A, De Guio R (2015) Modelling and uncertainties in integrated energy planning.
Renew Sustain Energy Rev 46:62–69. doi:10.1016/j.rser.2015.02.028
30. Refsgaard JC, Henriksen HJ (2004) Modelling guidelines––terminology and guiding princi-
ples. Advan Water Resour 27(1):71–82. doi:10.1016/j.advwatres.2003.08.006
31. Stachowiak H (1973) Allgemeine Modelltheorie. Springer, Berlin
32. Schneeweiß C (1992) Plannung 2, Konzepte der Prozeß- und Modellgestaltung, vol 2.
Springer, Berlin
33. Elert M, Butler A, Chen J, Dovlete C, Konoplev A, Golubenkov A, Sheppard M, Togawa O,
Zeevaert T (1998) Effects of model complexity on uncertainty estimates. J Environ Radioact
42(2–3):255–270. doi:10.1016/S0265-931X(98)00058-7
34. Pfenninger S, Hawkes A, Keirstead J (2014) Energy systems modeling for twenty-first cen-
tury energy challenges. Renew Sustain Energy Rev 33:74–86. doi:10.1016/j.rser.2014.02.003
35. Keirstead J, Jennings M, Sivakumar A (2012) A review of urban energy system models:
approaches, challenges and opportunities. Renew Sustain Energy Rev 16(6):3847–3866.
doi:10.1016/j.rser.2012.02.047
36. Tashakkori A, Teddlie C (1998) Mixed methodology : combining qualitative and quantitative
approaches. Sage, Thousand Oaks
37. Tashakkori A, Teddlie C (1998) Mixed methodology: combining qualitative and quantitative
approaches. Sage Publications, Thousand Oaks
38. Mingers J, Rosenhead J (2004) Problem structuring methods in action. Eur J Oper Res
152(3):530–554. doi:10.1016/s0377-2217(03)00056-0
39. Mingers J, Brocklesby J (1997) Multimethodology: towards a framework for mixing method-
ologies. Omega 25(5):489–509. doi:10.1016/S0305-0483(97)00018-2
40. Rosenhead J, Mingers J (2001) Rational analysis for a problematic world revisited: problem
structuring methods for complexity, uncertainty and conflict. Wiley, Chichester
References 23

41. Greene JC, Caracelli VJ, Graham WF (1989) Toward a conceptual frame-
work for mixed-method evaluation designs. Educ Eval Policy Anal 11(3):255–274.
doi:10.3102/01623737011003255
42. Molina-Azorίn JF (2011) The use and added value of mixed methods in management
research. J Mixed Methods Res 5:7–24. doi:10.1177/1558689810384490
43. Creswell JW (2009) Research design: qualitative, quantitative, and mixed methods
approaches. Sage Publications, Thoudand Oaks
44. Ascough Ii JC, Maier HR, Ravalico JK, Strudley MW (2008) Future research challenges for
incorporation of uncertainty in environmental and ecological decision-making. Ecol Modell
219(3–4):383–399. doi:10.1016/j.ecolmodel.2008.07.015
45. Asselt V (2000) Perspectives on uncertainty and risk, the PRIMA approach to decision sup-
port. Springer, US
46. Walker WE, Harremoës P, Rotmans J, van der Sluijs JP, van Asselt MBA, Janssen P, Krayer
von Krauss MP (2003) Defining uncertainty: a conceptual basis for uncertainty management
in model-based decision support. Integr Assess 4(1):5–17. doi:10.1076/iaij.4.1.5.16466
47. Refsgaard JC, van der Sluijs JP, Højberg AL, Vanrolleghem PA (2007) Uncertainty in the
environmental modelling process—a framework and guidance. Environ Modell Softw
22(11):1543–1556. doi:10.1016/j.envsoft.2007.02.004
48. Norton JP (2006) To what extent, and how, might uncertainty be defined? Integr Assess J
Bridging Sci Policy 6(1):83
49. Klauer B, Brown JD (2004) Conceptualising imperfect knowledge in public decision
making: ignorance, uncertainty, error and ‘risk situations’. Environ Res Eng Manage
27(1):124–128
50. Rommelfanger HJ, Eickemeier SH (2002) Entscheidungstheorie: klassische Konzepte und
Fuzzy-Erweiterungen; mit 109 Tabellen. Springer, Berlin
51. Joseph Y. Halpern (2005) Reasoning about uncertainty. University Press Group Limited
52. Weatherford R (1982) Philosophical foundations of probability theory. Routledge & K. Paul,
London
53. French S (1993) Decision theory: an introduction to the mathematics of rationality. Ellis
Horwood
54. Belohlavek R, Klir GJ, Lewis Iii HW, Way EC (2009) Concepts and fuzzy sets: misunder-
standings, misconceptions, and oversights. Int J Approximate Reasoning 51(1):23–34.
doi:10.1016/j.ijar.2009.06.012
55. Zadeh LA, Klir GJ, Yuan B (1996) Fuzzy sets, fuzzy logic, and fuzzy systems: selected
papers by Lotfi A. Zadeh. World Scientific Publishing Company, Incorporated
56. Zadeh LA (1965) Fuzzy set. Inform Control 8:338–353
57. Dubois DJ (1980) Fuzzy sets and systems: theory and applications. Elsevier Science,
Amsterdam
58. Chen G, Pham TT (2000) Introduction to fuzzy sets, fuzzy logic, and fuzzy control systems.
Taylor & Francis, UK
59. Zimmermann HJ (2001) Fuzzy set theory-and its applications. Kluwer Academic Publishers,
Dordrecht
60. Wang PP (1983) Advances in fuzzy sets, possibility theory, and applications. Plenum Press,
New York
61. Dubois D, Prade HM (1988) Possibility theory: an approach to computerized processing of
uncertainty. Plenum Press, New York
62. Klir GJ (1999) On fuzzy-set interpretation of possibility theory. Fuzzy Sets Syst 108(3):263–
273. doi:10.1016/S0165-0114(97)00371-0
63. Zadeh LA (1999) Fuzzy sets as a basis for a theory of possibility. Fuzzy Sets Syst 100
Supplement 1(0):9–34. doi:10.1016/S0165-0114(99)80004-9
64. Ross TJ, Booker JM, Parkinson WJ (2002) Fuzzy logic and probability applications: bridging
the gap. Society for industrial and applied mathematics (SIAM, 3600 Market Street, Floor 6,
Philadelphia, PA 19104)
24 1 Introduction

65. Rommelfanger H (1994) Fuzzy decision support-systeme. Springer, Berlin

66. Dubois D, Prade H (2012) Possibility theory. In: Meyers RA (ed) Computational complexity.
Springer, New York, pp 2240–2252. doi:10.1007/978-1-4614-1800-9_139
67. Dubois D (2006) Possibility theory and statistical reasoning. Comput Stat Data Anal
51(1):47–69. doi:10.1016/j.csda.2006.04.015
68. Dempster AP (1967) Upper and lower probabilities induced by a multivalued mapping. The
Ann Math Stat 38(2):325–339. doi:10.2307/2239146
69. Shafer G (1976) A mathematical theory of evidence. Princeton University Press, Princeton
70. Pearl J (1990) Reasoning with belief functions: an analysis of compatibility. Int J
Approximate Reasoning 4(5–6):363–389. doi:10.1016/0888-613X(90)90013-R
71. Srivastava R, Liu L (2003) Applications of belief functions in business decisions: a review.
Inform Syst Front 5(4):359–378. doi:10.1023/B:ISFI.0000005651.93751.4b
72. Helton JC, Johnson JD, Oberkampf WL (2004) An exploration of alternative approaches to
the representation of uncertainty in model predictions. Reliab Eng Syst Safety 85(1–3):39–
71. doi:10.1016/j.ress.2004.03.025
73. Zadeh LA (1984) Review of a mathematical theory of evidence. AI Mag 5(3):81
Chapter 2
Energy Infrastructure Planning
in Cities and Territories, Quality Factors
of Methods for Infrastructure Planning

The objectives of this chapter are

• Presentation of integrated energy planning in cities and territories (IEPCT),
­different phases, steps and general tasks
• Defining the energy infrastructure in cities and territories as a complex system
complex system
• Defining characteristic types of IEPCT Use Cases for implementation in the
­following chapters
• Describing the model building process by different model terms and according
to energy planning processes
• Providing the general requirements and quality factors of methods or method-
ologies supporting energy infrastructure planning in cities and territories

2.1 Introduction

The infrastructure in cities and territories is not only composed of a material com-
ponent, but it also involves social and economic components. It is impossible to
explain these components separately for long-range planning. Referring to critical
infrastructure systems, Ouyang [1] states “These systems are not alone but interde-
pendent at multiple levels to enhance their overall performance”. Considering the
whole, an infrastructure is the disposition of transportation, communications, fuel
and energy, water supply, and institutions in the fields of education, health, and
insurance [2]. Moreover, infrastructure in cities and territories needs to promote
eco-system integrity and environmental regeneration, thus avoiding environmen-
tal degradation and providing economic and social goods and services [3]. The
energy system in cities and territories is one of the critical elements of this infra-
structure. In this book, energy infrastructure will be the focus of the discussion

© Springer International Publishing Switzerland 2015 25

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_2
26 2 Energy Infrastructure Planning in Cities …

Table 2.1  Planning tasks and time horizon in energy industry

Planning category Time horizon Tasks and activities
Strategic planning More than 10 years Investment decisions for system replacement
or extension
Tactical planning 3 years Project risk management, …
1 years Budgeting, revisions, …
Operative planning 6 months Hydro reservoir planning, …
1 month Financial optimisation, …
1 week Procurement planning, …
1–3 days Optimisation, …
1 min Regulation power, …

with consideration of the interdependency with other infrastructure elements and

aspects.
Energy system planning by utility companies in a liberalized market can be
divided into three different categories according to the task or activities and the
time horizon. Table 2.1 shows the activities and time horizon in energy planning
based on the discussion in [4].
IEPCT, which is the focus in this book, refers to long-term strategic planning.
As stated in [5], “A local energy system consists of long-lived infrastructures
(a planning horizon of 10–30 years and eventually up to 50 years), which does not
lend itself to a quick modification or response.”
The next section of this chapter will present an energy planning procedure
based on a review that presents the tasks and activities systematically. Section 2.3
defines the energy system in cities and territories as a sociotechnical complex
infrastructure. In Sect. 2.4 two representative Use Cases are defined for implemen-
tation in the next chapters. Modelling and the model formalisation process in the
context of IEPCT are presented in Sect. 2.5. The overall requirements and quality
factors, which can be used for evaluation and design planning support methods,
are discussed in the Sect. 2.6. In the last section of this chapter, the main needs
and review outcomes are summarised.

2.2 Integrated Energy Planning in Cities and Territories

At the beginning of the nineties, several events lead to changes in the energy
market, and energy planning become more complex. These events include EU
directives for the liberalisation of the energy markets, increasing environmental
restrictions, an diverse interest of different people being involved in energy plan-
ning, a scarcity of fossil fuels, the use of intermitted renewable energies, and the
increasing share of small distribution generations systems.
In an extensive review preformed in a previous study [6], a generic IEPCT
framework was proposed for cities and territories. The planning is divided accord-
ingly into different phases and sub steps. The overall planning procedure is pre-
sented in Fig. 2.1.
2.2 Integrated Energy Planning in Cities and Territories 27

Fig. 2.1  General procedure of IEPCT (adapted from [6], used with permission of Elsevier)

From a methodological point of view, the planning processes are [6] divided
into the following four phases:
Phase I Preparation and orientations
Phase II Model design and detailed analysis
Phase III Prioritization and decision
Phase IV Implementation and monitoring
Different levels have been distinguished in the planning process [6]:
Participatory level, which describes planning participants
Process level, which describe planning tasks and activates
Methodological level, which shows methods and tools that are used to support dif-
ferent planning activities.

2.3 Energy Systems in City and Territory, a Sociotechnical

Infrastructure

An integrated energy system incorporates energy supply and demand systems

with many sub systems having different energy up- and downstream flows, ser-
vices from primary energy to final energy supply to the customer. The energy sup-
plier and user are not located anymore simply at the beginning and at the end of
energy chain, respectively. Indeed, even small final energy consumers might become
28 2 Energy Infrastructure Planning in Cities …

simultaneously electricity suppliers by using their own e.g. small PV system. The
intermittent nature of renewable energy resources, energy market deregulation, and
different interests and behaviour of different actors (supplier or energy consumer)
make planning and modelling of such complex integrated systems very challeng-
ing and requires the appropriate modelling paradigm taking into account uncertainty
issues. Integrated energy systems cannot be planned and modelled with only from the
technical (point of) view because aspects, such as behavioural issues of the energy
user or market requirements can have a strong influence on energy system design.
Urban or regional systems have different activities, natural processes, culture, trans-
portation or energy systems that are in different scales and have numerous interac-
tions that make the modelling task highly complex. Accordingly, an integrated energy
system in city or territory is a sociotechnical system. As defined in [7] “Sociotechnical
systems are systems that involve both complex physical–technical systems and net-
works of interdependent Actors”. It is a system of systems, an infrastructure involving
different technical manifestations and social organizations [8].
Long range energy system in cities or territories can’t be planned or analysed
focusing only on local conditions within certain administrative zone or by pure
geographic boundaries. Important parts of system like upstream flows of energy or
national energy price is not considered in this case which might have high impact
on planning and design of local system. How to consider this problem in planning
and modelling is discussed in the Chap. 3.

2.4 Defining Typology of Application or Use Cases

The methods discussed in this book concern integrated energy planning in cities
and territories (IEPCT) having more than 50,000 inhabitants for time horizons
greater than 10 years. However, the planning process conditions can differ. In
some situations, at the very beginning, an integrated model and plan is intended to
be developed using one quantitative overall model, which is developed by a par-
ticular planning group. In another situation, the planning can begin with an inte-
grated view for the city or territory development, without the possibility, or even
the need, for developing an integrated single quantitative model. Separate models
might be developed by different groups in different planning or project develop-
ment time spans. It is also uncertain whether these separate models will be com-
bined into one integrated, quantitative model in the near future. This leads us to
define two general type of Use Case I and II.

2.4.1 Use Case I: Decentralised Multi-model Based IEPCT

In this case, the integrated plan is not implemented in one integrated quantitative
model. Instead, different models might be developed by different teams separately
2.4 Defining Typology of Application or Use Cases 29

with large project time lag. One model, for instance, might refer to the transporta-
tion sector, while another might refer to heating systems. These models are not
integrated in one quantitative integrated model at the beginning of study and it is
not intended to do so in some project time frame.

2.4.2 Use Case II: Integrated-Model Based IEPCT

Here, planning is based on one integrated model, which is developed in the fore-
seeable future, in a given planning time frame by a certain planning and modelling
team. The integrated model can include different sub-models that are integrated
in one model having different quantitative links e.g., a model for transportation, a
cooling system or energy demand in industry.
Some mixtures of considered Use Cases are also possible but will not be con-
sidered here as separate cases.

2.5 Modelling in IEPCT

2.5.1 Models and Different Degrees of Formalisation

The modelling process can be divided into different degrees of formalization

that are linked to the planning process. Similar to planning modelling process
have different tasks and sub tasks which might be linked to the planning process
discussed in the Sect. 2.2.
Following discussion and terminology definitions are provided here based on
[9]. However, detailed information the interested reader can find in this reference.
Mental models are subjective, abstracted details of a specific world view of an
individual mental world [10]. They exist only in the mental world of planning par-
ticipants and are not interpersonal verifiable.
Conceptual models take into account the essential and relevant system elements
and interactions for the specific problem and goal, descriptions of flows processes,
system boundaries, aggregation level of analysis (temporal, spatial etc.), and a
time horizon without mathematical description at this stage of modelling.
Between these two models are the expressed models, which are explicit
descriptions of mental models. Mental models are expressed mostly in a verbal
form.
Formal models (or site-specific model in [11]) are a numerical description of
conceptual models that take into account natural laws, engineering, social and
economic interactions. It includes all parameters, variables and their values.
Formal models can not only be based on mathematical programming paradigms
but also on other paradigms, such as agent-based simulation.
Computer models are encoded models in a computer program.
30 2 Energy Infrastructure Planning in Cities …

Calibrated models are the verified, corroborated models evaluated and tested
by using different approaches, such as sensitivity or uncertainty analysis or cali-
brating some model parameters or functions against independent, exogenous
data.
Applied models are implemented to support a decision maker’s needs, such as
analysis scenarios and assessment impacts of actions. These models present an
opportunity for post auditing of the model: the model can be redesigned according
to the decision maker’s needs.
The link of different modelling and planning steps are presented in [9]. One
of the main output of planning phase I is accordingly the conceptual model,
which takes into account different study restrictions and planning participants
needs. However, the conceptual model can be updated in future planning stages
progressively. The formal model is usually developed in planning phase II and
III. The computer model is usually not separate from the formal model and it
developed simultaneously by implementing the formal model in computer code.
The applied or final developed model ready in planning phase III. However, the
modelling process continues in planning phase IV where the integrated plan
is fleshed out into several programs and subproject. The integrated model can
be divided into several partial models that support the planning activities in
phase IV. The interrelation of modelling and planning activities are presented in
Fig. 2.2.

Fig. 2.2  IEPCT process and modelling steps (adapted from [9] used with permission of Else-
vier)
2.6 Overall Requirements and Quality … 31

2.6 Overall Requirements and Quality Factors

of Energy Planning and Modelling Methods

A quality (from Latin qualities) is defined in [12] as “The standard of sth when
it is compared to other things like it; how good or bad sth is […]”. ISO 9000
defines quality as the “degree to which a set of inherent characteristics ful-
fils requirements”. And requirement is defined as “need or expectation that is
stated, generally implied or obligatory”. ISO 9000 definitions will be imple-
mented here. Certain requirements can have several characteristics or quality
factors which measure fulfilment of requirement, such as when a requirement
is “use less resources” for the implementation of a certain method. The measur-
able quality factors may be the “time” or “cost” required for implementing the
method. Requirements and quality factors can be organised together in a hierar-
chical manner as shown in Fig. 2.3 and Table 2.2, where quality factors belong
to the lowest level of measurable indicators. The structure and composition of
the requirement tree can vary depending on study needs. As there are no pre-
defined IEPCT planning standards, regulations or broth consensus known yet, the
requirements and quality factors are defined based on a review of the empirical
study’s needs.
The overall quality of planning and modelling depends on several factors, such
as implemented methods, experiences and competencies of planning participants,
and conformity of the planning and decision-making procedure with existing laws.
The planning atmosphere/venue can also play an important role. The quality of
planning and modelling results/outputs are also a part of the overall planning qual-
ity. Planning and modelling under uncertainty, where planning output cannot be
evaluated exactly, the planning process and used methods or methodology becomes
the focus for quality evaluation [13]. These general factors provide the opportunity
to see the extent of different aspects influencing the entire planning quality.

Fig. 2.3  Overall planning quality factors

32 2 Energy Infrastructure Planning in Cities …

Table 2.2  Requirements and quality factors of planning methods

Requirement Quality factor References
Technical considerations Level of validity and legitimacy [20–24]
Generation of required outputs See Chaps. 3 and 4
Holistic [4, 21, 23–31]
Incorporation of qualitative [32]
and quantitative information
Incorporation of different types See Chap. 3
of uncertainty
Organisational capability Required time, money, expertise [11, 24, 26]
Flexibility [27, 33–35]
Satisfaction by planning Level of satisfaction with method [20, 23, 24, 32, 34, 36]
participants Level of satisfaction with results
Knowledge discovering and Level of avoidance of mental [24, 25, 27, 37–43]
learning inertia and learning
Capability to identify and solve
contradiction
Collaboration support Level of identification of conflicting [24–26, 30, 33, 37,
interests 44–46]
Level of conflict resolution support
Level of interaction

A framework for the evaluation of overall decision quality in the context of

territory planning and management for resolving geographic boundary c­ onflicts,
not only for energy related tasks, is suggested by [14]. Quality of planning is
divided into two major groups: quality of the decision process and quality of the
decision outcome as shown in Fig. 2.3. The quality of the decision process is
divided into three sub categories: quality of the decision procedure, quality of the
decision unit (planning participants) and quality of the decision method [14].
The requirements and quality factors of planning and decision support meth-
ods suggested by [14] are examined in the context of IEP in cities and territories
here. The need for additional quality factors is identified by reviewing the energy
planning and modelling studies. The review has been performed using journals,
such as Applied Energy, Energy Studies Review, International Journal of Energy
Research, Sustainable Cities and Society, Renewable and Sustainable Energy
Reviews, Energy, Energy for Sustainable Development, Energy Policy or Urban
Studies. The keywords used in the review were quality factor or criteria, ­property,
features, requirements, and proposed function of method. Books and technical
reports were reviewed as well [15–19]. Very few studies discuss explicitly the
requirements or quality factors of methods for IEP in cites or territories. However,
it was possible to collect some factors (Table 2.2). The factors are structured hier-
archically. The quality factors can be evaluated either quantitatively or qualita-
tively. In Chap. 5, these quality factors are specified and adapted for uncertainty
analysis methods.
2.6 Overall Requirements and Quality … 33

In general for rational choose, which is also the case for IEP in cities and ter-
ritories three elements are important:
What do I want?—Goals and values of decision maker and planning participants
What can I do?—Initiatives, strategies or solutions to reach goals
What might happen?—Outcomes: Situations today and in the future.

Technical consideration
Technical consideration is the main functional requirement of methods or meth-
odology implemented in planning. Three different quality factors can be used for
classification this requirement.
Validity and legitimacy of method depends on different aspects. E.g. conform-
ity of implemented method can be dependent on national or international norms
like ISO-norms. This factor has formal character evaluating whether the method is
conform to existing regulation or norms.
Generation of required outputs characterise whether the method or methodol-
ogy produce required output for given planning and modelling step fully or par-
tially. Let’s notice that among others, some methods have to support generation of
objectives and participants values, other methods have to provide numerical values
of parameters of solution models, or outline the types and levels of uncertainty.
Holistic aspect refers in IEPCT studies the consideration of different aspect like
economic, environmental or technical aspects, whole energy infrastructures tak-
ing into account different preferences of decision maker. Holistic view is particu-
larly important for integrated studies for supporting long term sustainable territory
development because it measures the possibility of method to appropriately incor-
porate different aspects and elements of infrastructure.
Incorporate qualitative and quantitative information: In long term interactive
planning available information can be in quantitative e.g. statistical survey or qual-
itative like expert judgment, also preferences of decision maker, which are qualita-
tive, need to be included in the analysis.
Incorporation of uncertainty is factor which shows whether all uncertainty
aspect related to planning are considered and to what extent. These aspects are
discussed in Chap. 5 more detailed.
Organisational capability
Organisational capability presents the aspects of flexibility and required resources
for using the method or methodology.
Required time, money, expertise indicates how much time, resources or exper-
tise is required for implementation of the method.
Flexibility of methods or methodology refers to whether it can be imple-
mented and adapted according to different planning and modelling situations,
Use Cases or phases without large effort e.g., using method or methodology in
other conditions which requires less data, resources or expertise without large
modification.
34 2 Energy Infrastructure Planning in Cities …

Knowledge discovering and learning

Level of avoidance of mental inertia and learning is an important factor that shows
whether the method can help avoid mental inertia and support learning processes
for the decision maker.
Identify and solve contradiction is another important aspect mentioned in some
studies. Different categories of contradictions exist.
Satisfaction by planning participants
This factor reflects the decision maker’s positive evaluation of the method. It is
closely related to factor legitimacy or credibility discussed above.
Satisfaction with methods is a factor that can be evaluated directly in an empiri-
cal study testing or by using the method in real time. It measures the level of satis-
faction with the method. Usually it depends on familiarity using the method.
Satisfaction with results is a factor indicating the level of satisfaction of method
results by planning participants.
Collaboration support
Conflict identification and resolution is particularly important in an interactive
planning situation with different stakeholders and planning participants. It deter-
mines whether the method can help identify conflicted interests and explicitly
­support conflict resolution.
Level of interaction is particularly important in group decision making as iden-
tified by [47]. The method has to support collaborative, fair and open decisions as
opposed to manipulative decisions [14].
Quality factors using different methods together
The quality factors discussed previously refer to single methods. However, using
different methods together for different planning tasks or phases is common.
Therefore, additional aspects and rationales from mixed method research will be
considered and discussed in Chap. 5 for designing a multi-method approach for
uncertainty analysis.

2.7 Summary and Open Problems

The main conclusions of this chapter can be summarised as follow:

• The IEP in cities and territories is a complex, multi stage task that needs to take
into account different aspects, such as environmental, technical or economic,
different participants’ interests and a complex, open energy infrastructure
• There are several tasks that have not been addressed systematically using meth-
ods or methodology yet, such as integrated modelling or uncertainty analysis.
These issues will be discussed in greater detail in Chaps. 3 and 5
2.7 Summary and Open Problems 35

• Studies show that IEPCT in a conditions where no probability information

about the states of environment is available or only some events probability can
be roughly defined. IEPCT is performed in uncertain environment according to
definition in Sect. 1.3.6
• IEPCT is a group decision-making procedure
• The planning and selection of plan has to be done taking into account multiple
criteria in limited budget conditions
• IEPST is not flexible planning as defined in Sect. 1.3.6. It is rather rigorous
planning for next 10–30 years, which does not lend itself to quick modification
or response.
• None of the reviewed method can address all of the planning needs
• There are multiple requirements and quality factors which have to be consid-
ering when evaluating or developing of methods or methodologies to support
IEPCT.

References

1. Ouyang M (2014) Review on modeling and simulation of interdependent critical infrastruc-

ture systems. Reliab Eng Syst Saf 121:43–60. doi:10.1016/j.ress.2013.06.040
2. Fiedorowicz K, Rzepka G (1977) Planning of the economic infrastructure. East Eur Econ
16(1)
3. Bond P (1999) Basic infrastructure for socio-economic development, environmental
protection and geographical desegregation: South Africa’s unmet challenge. Geoforum
30(1):43–59. doi:10.1016/S0016-7185(98)00031-1
4. Makkonen S (2005) Decision modelling tools for utilities in the deregulated energy market.
Helsinki University of Technology
5. Bhatt V, Friley P, Lee J (2010) Integrated energy and environmental systems
analysis methodology for achieving low carbon cities. J Renew Sustain Energy 2(3).
doi:10.1063/1.3456367
6. Mirakyan A, De Guio R (2013) Integrated energy planning in cities and t­erritories:
a review of methods and tools. Renew Sustain Energy Rev 22:289–297.
doi:10.1016/j.rser.2013.01.033
7. de Bruijn H, Herder PM (2009) System and actor perspectives on sociotechnical
systems. IEEE Trans Syst Man Cybern Part A Syst Hum 39(5):981–992. doi:10.1109/
tsmca.2009.2025452
8. Hansman RJ, Christopher M, De Richard N, Renee R, Daniel R (2006) Research agenda
for an integrated approach to infrastructure planning, design and management. IJCIS
2(2–3):146–159. doi:10.1504/ijcis.2006.009434
9. Mirakyan A, De Guio R (2015) Modelling and uncertainties in integrated energy planning.
Renew Sustain Energy Rev 46:62–69. doi:10.1016/j.rser.2015.02.028
10. Zahn E (1991) Strategieunterstützngssysteme. In: Milling P (ed) Systemmanagement und
Managementsysteme. Duncker & Humblot, Berlin
11. Refsgaard JC, Henriksen HJ (2004) Modelling guidelines—terminology and guiding princi-
ples. Adv Water Resour 27(1):71–82. doi:10.1016/j.advwatres.2003.08.006
12. Hornby AS (2000) Oxford advanced learner’s dictionary. Oxford University Press, Oxford
13. Keren G, de Bruin WB (2005) On the assessment of decision quality: considerations regard-
ing utility, conflict and accountability. In: Thinking: psychological perspectives on reasoning,
judgment and decision making. Wiley, New York, pp 347–363. doi:10.1002/047001332X.ch16
36 2 Energy Infrastructure Planning in Cities …

14. Kamruzzaman M (2007) Enhancing the quality of decision making? In: Introducing spatial
multi criteria evaluation for boundary conflict resolution in the Philippines. International
Insitute for Geo-information Science and Earth Observation, The Netherlands
15. Allen PM (1997) Cities and regions as self-organizing systems: models of complexity.
Gordon and Breach Science Publishers
16. Flavin C (1984) Electricity’s future: the shift to efficiency and small-scale power, 61.
Worldwatch Institute, Washington, DC
17. Forrester JW (1970) Urban dynamics. MIT Press, Cambridge (3 print/XIII)
18. Brief diagnosis of the energy and environment context of Singapore (2010) EIFER-Technical
report, Karlsruhe
19. Schmidt S et al (2012) Mexico waste to energy model. Technical report, EIFER, Karlsruhe
20. Wene C-O, Rydén B (1989) A comprehensive energy model in the municipal energy
planning process. Math Comput Model 12(8):1050. doi:10.1016/0895-7177(89)90223-9
21. Gunnarsson-ÖStling U, HÖJer M (2011) Scenario planning for sustainability in Stockholm,
Sweden: environmental justice considerations. Int J Urban Reg Res 35(5):1048–1067.
doi:10.1111/j.1468-2427.2010.01002.x
22. Cai YP, Huang GH, Yang ZF, Tan Q (2009) Identification of optimal strategies for energy
management systems planning under multiple uncertainties. Appl Energy 86(4):480–495.
doi:10.1016/j.apenergy.2008.09.025
23. Løken E (2007) Use of multicriteria decision analysis methods for energy planning prob-
lems. Renew Sustain Energy Rev 11(7):1584–1595. doi:10.1016/j.rser.2005.11.005
24. IEA (2000) ANNEX 33. Adv Local Energy Plan Methodol (ALEP), A Guidebook
25. Fielden D, Jacques JK (1998) Systemic approach to energy rationalisation
in island communities. Int J Energy Res 22(2):107–129. doi:10.1002/(sici)1099-114x
(199802)22:2<107:aid-er289>3.0.co;2-u
26. Rotmans J, Van Asselt MBA (2000) Towards an integrated approach for sustainable city
planning. J Multi-Criteria Decis Anal 9(1–3):110–124. doi:10.1002/1099-1360(200001/05)9:
1/3<110:aid-mcda270>3.0.co;2-f
27. Cormio C, Dicorato M, Minoia A, Trovato M (2003) A regional energy planning
methodology including renewable energy sources and environmental constraints. Renew
Sustain Energy Rev 7(2):99–130. doi:10.1016/S1364-0321(03)00004-2
28. Yin Y, Cohen S, Huang GH (2000) Global climate change and regional sustainable
development: the case of Mackenzie basin in Canada. Integr Assess 21(26)
29. Sarafidis Y, Diakoulaki D, Papayannakis L, Zervos A (1999) A regional planning approach
for the promotion of renewable energies. Renew Energy 18(3):317–330
30. Williams PM (2002) Community strategies: mainstreaming sustainable development and
strategic planning? Sustain Dev 10(4):197–205. doi:10.1002/sd.197
31. Ivner J (2009) Do decision-making tools lead to better energy planning?. Linköping
University, Linköping
32. Polatidis H, Haralambopoulos DA, Munda G, Vreeker R (2006) Selecting an appropriate
multi-criteria decision analysis technique for renewable energy planning. Energy Sour Part B
Econ Plan Policy 1(2):181–193. doi:10.1080/009083190881607
33. Neves LMP, Martins AG, Antunes CH (2004) Using SSM to rethink the analysis of energy
efficiency initiatives. J Oper Res Soc 55(9):968–975. doi:10.1057/palgrave.jors.2601763
34. Ramachandra TV (2009) RIEP: regional integrated energy plan. Renew Sustain Energy Rev
13(2):285–317. doi:10.1016/j.rser.2007.10.004
35. Graeber B, Spalding-Fecher DR (2000) Regional integrated resource planning and its role in
regional electricity co-operation and development in Southern Africa. Energy Sustain Dev
4(2):32–37. doi:10.1016/S0973-0826(08)60240-9
36. Lam HL, Varbanov PS, Klemes JJ (2011) Regional renewable energy and resource planning.
Appl Energy 88(2):545–550. doi:10.1016/j.apenergy.2010.05.019
37. Coelho D, Antunes CH, Martins AG (2010) Using SSM for structuring decision support in
urban energy planning. Technol Econ Dev Econ 16(4):641–653. doi:10.3846/tede.2010.39
References 37

38. Mirakyan A, Lelait L, Khomenko N, Kaikov I (2009) Methodological framework for the
analysis and development of a sustainable, integrated, regional energy plan—a French region
case study. In: Paper presented at the EcoMod, Ottawa
39. Butera FM (1998) Moving towards municipal energy planning—the case of Palermo:
the importance of non-technical issues. Renew Energy 15(1–4):349–355. doi:10.1016/
S0960-1481(98)00185-2
40. Trutnevyte E, Stauffacher M, Scholz RW (2012) Linking stakeholder visions with resource
allocation scenarios and multi-criteria assessment. Eur J Oper Res 219(3):762–772.
doi:10.1016/j.ejor.2012.01.009
41. Rotmans J, van Asselt M, Vellinga P (2000) An integrated planning tool for sustainable cities.
Environ Impact Assess Rev 20(3):265–276
42. Glasson J (1992) The fall and rise of regional planning in the economically advanced nations.
Urban Stud 29(3–4):505–531. doi:10.1080/00420989220080551
43. Mirakyan A, Khomenko N, Lelait L, Kaikov I (2009) The potential of OTSM-TRIZ as a
frameworking method for modern regional, integrated energy planning and modeling. In:
Paper presented at the the fifth TRIZ symposium in Japan, Tokyo
44. Terrados J, Almonacid G, Hontoria L (2007) Regional energy planning through SWOT anal-
ysis and strategic planning tools.: impact on renewables development. Renew Sustain Energy
Rev 11(6):1275–1287. doi:10.1016/j.rser.2005.08.003
45. Nicholas Lenssen (1996) Local integrated resource planning: a new tool for a competitive
era. Electr J 9(6):26–36
46. Linkov I, Varghese A, Jamil S, Seager T, Kiker G, Bridges T (2005) Multi-criteria decision
analysis: a framework for structuring remedial decisions at contaminated sites comparative
risk assessment and environmental decision making. In: Linkov I, Ramadan A (eds) NATO
science series IV: earth and environmental sciences, vol 38. Springer, Netherlands, pp 15–54.
doi:10.1007/1-4020-2243-3_2
47. Timmermans D, Vlek C (1996) Effects on decision quality of supporting multi-attribute
evaluation in groups. Organ Behav Hum Decis Process 68(2):158–170. doi:10.1006/
obhd.1996.0096
Chapter 3
3-Domain Modelling

The objectives of this chapter are

• Presentation of a 3-domain Metasystem framework, which allows the identification
of appropriate modelling approaches while considering not only the system but also
the multiple study perspectives and planning constraints such as limited expertise or
data.
• Discussion of the 3-domain modelling concept, an integrated concept of differ-
ent modelling methods used in different domains
• Providing a mapping between different modelling approaches and different
domains and use cases on the basis of literature discussions and experiences.

3.1 Introduction

After developing a conceptual model there are still questions remaining to develop
formal and other models. Among them:
• How can the model complexity of energy infrastructure be reduced while con-
sidering different planning perspectives? As discussed in Sect. 2.3, the energy
system is a social-technical complex infrastructure without clear boundaries,
which might evoke different uncertainties?
• How are the amount of resources and time requirements in model development
reduced while maintaining model adequacy for planning purposes and the
modelled system at the same time?
• How is the limited availability of expertise or knowledge dealt with, at least
explicitly, for planning of the overall complex infrastructure?
• How is the model of the overall system represented according to the interests and
responsibility of decision makers and planning participants?

© Springer International Publishing Switzerland 2015 39

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_3
40 3 3-Domain Modelling

To address the above questions, 3-domain Modelling concept is proposed in this

chapter, which is based on the scheme of the 3-domain Metasystem discussed in
Sect. 3.2. Lorenz and Jost [1] argue that finding the most appropriate modelling
method rests on the clarification of the object to be modelled and the modelling
purpose. With the 3-domain Metasystem framework, the different needs for each
domain are identified and existing modelling approaches are reviewed accord-
ing to these needs in Sect. 3.3. The appropriateness of an individual modelling
approach is based either on the literature discussion or on the implementation and
empirical analysis provided in Chap. 6. In Sect. 3.4.4.1 a new approach for model-
ling a distant domain is proposed.

3.2 3-Domain Metasystem

The central question in the system analysis is “how to define a system?” Klir [2]
provides a hierarchy of epistemological levels of systems for development of
methodologies or organization of a set of methods for systems analysis and prob-
lem solving. Epistemological levels are distinguished from each other by the level
of additional knowledge regarding the variables of the associated system, which is
not available at the lower level. Five system epistemological levels are presented
in [2] based on the interaction between the investigator, object and environment.
The lowest ‘0’ level is defined as the ‘source system’, and it is the system on the
conceptual level. The system is described via a set of variables and their potential
states. It is denoted as ‘source system’ to indicate that such a system is, at least
potentially, a source of empirical data. The system is defined according to the
study propose and investigation constraints (e.g., time and resources). Each higher
level of a system includes all of the knowledge of any lower level systems and
contains some additional knowledge, which is not available at the lower levels [2].
The highest level of a system, ‘4’ or ‘5’, contains all of the knowledge regarding
the corresponding lower level systems. This higher level system is defined as a
‘metasystem’. It contains information not only about the data and their potential
states but also a metacharacterization of knowledge such as rules, relations or pro-
cedures. If interested, a reader can find a detailed discussion about different levels
of a system and their characterization in [2].
Some notion of this approach will be implemented for the design of the
3-domain Metasystems as a framework for identification and development of mod-
elling approaches.
To avoid having ‘considered a model’ and ‘not considered an environment’
in modelling, which is common praxis in modelling studies, a 3-domain Meta-
system framework of the overall system is proposed here (Fig. 3.1). According to
a particular studies needs and system considerations, the overall system is con-
sidered in three domains. It is important to note that boundaries among domains
may not necessarily correspond to defined physical or spatial demarcation such as
the spatial boundaries of a region. Indeed, when the system is considered from
3.2 3-Domain Metasystem 41

Distant
domain Neighbouring
domain

Targeted
domain

Causal forces
Driving forces

Fig. 3.1  Metasystem of modelling domains

an economic or/and administrative point(s) of view, a domain may not have a

particular spatial association. For example, in a life cycle assessment of energy
production, the upstream processes are often located in several administrative or
economic zones. Likewise, a domain is not always strongly linked with certain
planning branches such as the energy system. Rather, it can be defined according
to the specific needs of the study. For instance, in the case study of Mexico City
(see Chap. 6), the targeted domain is “waste and energy” infrastructure together.
The domains can have a different epistemological level according to [2]. Whereas
for the targeted domain, a high knowledge might be available and is required, for
a neighbouring or distant domain, a high epistemological level is not required.
Therefore, a neighbouring and distant domain can be considered on the level of
a ‘data system’ or ‘generative system’ defined by [2], which means a system image
can not only be characterised by variables with a function of their actual states but
also their potential future states, at least in qualitative terms. In contrast, the targeted
domain can be considered on a higher epistemological level ‘structure system’, which
is additionally defined in terms of a set of generative systems, which are referred to
as subsystems of the overall system. The 3-domain Metasystem we consider a ‘meta
system’ with some metacharacterization (rules, relations, and procedures) [2].
The targeted domain equals the domain where problems, goals and possible solu-
tions are located. When the planners or modellers have certain domain knowledge
of this area; they can develop and manipulate causal models of this domain.
Usually, this is the domain where a decision maker has certain power to imple-
ment actions and can at least partially control or design the system. Planning par-
ticipants are interested in the structure of the system and processes. They have
more explanatory expectations or system analysis in time and space. In addition,
they are interested in causation and the description of a systems current and future
42 3 3-Domain Modelling

possible states. In this domain, not only foresight models but also impact and
decision models are required (see model definitions in Sect. 1.3.3).
Neighbouring domain: Although problems and goals are not located in this domain
(e.g., demography, which might not be the primary problem in some energy plan-
ning contexts), there might be some important drivers in a neighbouring domain that
have a high influence on the targeted domain models and also have a strong interre-
lation. Planning participants do not have enough knowledge or resources to analyse
and model the system structure and processes (e.g., demographic) in this domain.
However, they might be more interested by having a description of system states
using foresight models and less interested for causation, impact or decision models.
Distant domain: Usually, local planners or modellers have less experience or knowl-
edge of distant domains and there might be no real need to develop and analyse
system structure and processes in a distant domain. This domain can be related to
international or transnational issues, such as the distant domain of local energy plan-
ning but also to some aspect inside the city or territory, such as crime, which might
be not directly linked with energy planning problems or goals. However, planners or
decision makers can be interested in general information, such as economic trends
and cycles, or large changes such as climate change or migration effects. For the
distant domain, planning participants can be interested in the description of system
states and less interested in causation or decision models because they might have
neither competencies nor responsibilities for making decisions in this domain.
In some situations, it might be useful to integrate the neighbouring and distant
domain as a whole instead of considering them separately.
Causal forces are forces from the distant domain that can impact models in a
neighbouring or targeted domain.
Driving forces are forces from a neighbouring domain that can impact models in a
targeted domain.
The opposite direction of influence from targeted to neighbouring domain also
exist, and in the sense of identification of the main drivers in the neighbouring
domain, which may have a higher influence on the targeted domain.
The targeted domain has a passive role on the distant domain in a city and
territory modelling case. The impacts of local planning on the targeted domain and
on global issues in the distant domain are incorporated quantitatively rather than
on a higher level such as national level planning.
Each domain needs to be ‘identifiable’ and ‘consistent’ in terms of the system
requirement and planning problems. The integration of domains needs to satisfy
several requirements. Klir [2] provides different requirements when i­ntegrating
several systems into a large system. All 3-domains and their interactions are
defined here as a Metasystem.
Defining Ax as a sets of variables of particular x sub systems in a given domain,
A will be defined as a total amount of variables of each domain.
d

Ad = Ax
x∈N
3.2 3-Domain Metasystem 43

If two domains are coupled, then they share coupling variables, which are defined
above as casual or driving forces. If two different domains, e.g., targeted domain
Atx and neighbouring domain Anx share variables, then
Atx ∩ Anx �= ∅
Sharing variables of different domains must be compatible and have the same
type of data with the same units, etc. None of the variables in the sets of Ax of the
domains, as defined above, is allowed to be declared as an output variable in more
than one of the domains to keep consistency among the domains. Another required
condition is that the shared variables in each domain need to have the same pro-
jected behaviour. Assuming that S x is the sampling variable of the Ad in the set of
variables, then
Ad ⊇ S x
The behaviour consistency conditions will ensure that the behaviour functions fBt
and fBn for each pair of domains, with respect to variables they share (coupling
variables), are equal.
fBt ↓ Sxt ∩ Sxn = fBn ↓ Sxn ∩ Sxt
However, coupling variables must not be computed synchronically. This is particu-
larly required when the coupling variable has different scales in different models
and there is a need for adapting scales before synchronisation.
The proposed 3-domain Metasystem should allow the representation of the rel-
evant planning system in a holistic, coherent and pragmatic way while maintaining
the parsimony principle. The 3-domain Metasystem will also support the balance
between complexity, data availability, planning perspective and uncertainty trades
off. Two general principles of system complexity are identified by Ayyub [3],
which need to be considered:
• The measure of the complexity of a system should be proportional to the
amount of information required to describe the system.
• The measure of the complexity of a system should be proportional to the
amount of information needed to resolve any uncertainty associated with the
system involved.

3.3 3-Domain Modelling: Different Approaches

for Different Domains

3.3.1 Introduction

The 3-domain Metasystem discussed above provides more flexibility for model
formalisation according to the available information, expertise and study needs.
The focus of the discussion below is concentrated on foresight and impact model-
ling approaches (see definition in Sect. 1.3.3).
44 3 3-Domain Modelling

Three types of modelling approaches, according to the information source used,

are discussed in this section and then linked to the 3-domain Metasystem notion
presented before.
The information sources used in the planning and modelling may range from the
‘hard’ to the ‘soft’ spectrum. Ford [4] proposed an information spectrum starting with
information that is based on well specified physical laws and experiments followed by
social system data, expert judgment and ending by personal intuition with very few
formal specification in the end of the spectrum. Whereas ‘hard’ information, based
on laws or experiments delivers usually numerical data, the ‘soft’ information source,
based on judgment or intuition might provide only general qualitative information.
However, ‘soft’ information source data also plays an important role for long-range
planning and modelling because the information availability is low and expertise is
required. Three main categories of modelling approaches have been defined in [5],
which can be considered a grouping of the spectrum of information sources discussed
before. These modelling approaches are named: data-driven modelling, process-driven
modelling and judgmental-driven modelling. Process-driven modelling is based on
laws and axioms (such as physical or environmental). Data-driven modelling uses
information sources based on statistical surveys or measurements implementing in
statistical approaches or in machine learning. Judgmental-driven modelling is based
on judgments of experts, intuitions of experts or planning participants. There are cor-
respondence between these categories and the definitions of modelling terminologies
discussed in the Sect. 1.3.3. For instance, data driven modelling approaches are quan-
titative simulation modelling approaches, while judgmental modelling approaches are
qualitative system modelling approaches, and process driven modelling approaches
can be a quantitative and/or qualitative system or simulation modelling approaches.
These approaches can be used in combination such as data-driven and judgmental-
driven approaches. However, each approach can provide a superior contribution in
combination with other methods for certain modelling domains discussed above.
The modelling approaches discussed here are deterministic. Modelling stochas-
ticity and analysing of uncertainty are discussed in Chap. 5.

3.3.2 Data-Driven Modelling

This type of modelling approach is basically data oriented and based on surveys
or measurements. Data driven modelling is not oriented on or specific to a certain
branch. For example, data driven modelling does not distinguish whether the data
describe demography or financial issues. It considers only the data behaviour.
Methods that are used for this type of modelling are based, e.g., on univariate mod-
els, such as time series mathematical methods, or on multivariate models, such as
neural networks, which can also be used for univariate analysis. This type of model-
ling can be employed for historical or current data descriptive analysis and inductive
analysis based on given historical data behaviour. Data-driven modelling requires the
least contextual information for modelling of a given planning object.
3.3 3-Domain Modelling: Different Approaches for Different Domains 45

Numerous methods have been implemented and tested in different studies.

A large initiative implementing and comparing diverse data driven time series
analysis methods for forecasting is the M-competition. In this initiative, different
methods are tested, and their performances [6–8] are compared.
Some semi-quantitative methods for pattern recognition, such as S-shaped
curve methods, which are based not only on the available data but also on analo-
gies, can be considered here as well. These types of methods can be implemented,
e.g., for modelling technology diffusion issues in the neighbouring domain. The
selection of some methods for implementing in neighbouring or distant domains is
performed below in Sect. 3.4.3.

3.3.3 Process-Driven Modelling

Process-driven modelling approaches are primarily process oriented, based on

contextual knowledge about, e.g., energy or economic laws and theories. Different
modelling approaches, such as economic or physical (like energy) equilibrium
modelling or system dynamics modelling are used for this purpose. These models
are mostly quantitative simulation models based on a set of equations.
Choosing a modelling method depends on many aspects. A review by [9] noted
that most of IEPCT studies have used physical or economic equilibrium models.
Out of 35 studies, only two of them used complex system modelling approaches in
IEPCT. However, Bale et al. [10] indicated different limitations of widely used mod-
elling approaches, such as not considering the heterogeneity of system actors or mar-
ket imperfection and emergency. They argue that understanding the energy system
change would benefit from the application of complexity science thinking and mod-
elling such as system dynamics. Additionally, as stated in Sect. 2.3 and by Albeverio
et al. [11] urban or regional systems are a complex social and technical infrastructure
having different activities, natural processes, cultures, transportation or energy sys-
tems and multiple interactions on different scales: time and space. This makes plan-
ning and modelling tasks very complex and requires an adequate system modelling
approach. A combination of different approaches using different abstraction levels or
formalisation is also possible for implementation in this domain [12]. The key point
in this modelling category is that it relays on explicitly specified causation, which is
based on certain physical and economic laws, according to certain theory.

3.3.3.1 Comparison of Complex System Modelling Approaches

The range of complex system modelling approaches varies from cell based
behavioural modelling to self-organised evolutionary complex system model-
ling approaches. Comparing different complex system modelling approaches [13]
concludes “[…] in general using Agent Based modelling approach you are able
to capture more real life phenomena than with system dynamics or discrete event
46 3 3-Domain Modelling

simulation approach.” However, modelling approaches depend not only on the

modelling object—the real system—but also on the modelling perspective or study
objective, available information and skills of the modeller or planner. A list of con-
ditions indicating when the implantation of AB modelling is more appropriate is
provided by [14]. Accordingly, AB has to be implemented:
• When the interactions between the agents are complex, nonlinear, discontinuous
or discrete,
• When space is crucial and the agents’ positions are not fixed,
• When the population is heterogeneous and when each individual is (potentially)
different,
• When the topology of interactions is heterogeneous and complex, and
• When the agents exhibit complex behaviour, including learning and adaptation.
Agent Based modelling requires a much higher effort for validation and verifica-
tion of the model (model evaluation), for instance [15]. Furthermore, there must be
enough information to specify each heterogeneous agent. As stated by [16] “The
complexity of a model must be weighed against the availability and uncertainty
of the input data because the apparent advantage of using a more complex model,
which may give a better representation of reality, may be overshadowed by the
introduction of greater data uncertainty”.
Diverse and complex system modelling approaches are discussed in [12] and
[13]. When it is important to analyse discrete events at a detailed level, then dis-
crete event simulation (DES) is appropriated, particularly for short-term operative
planning tasks. In contrast, SD aggregates different events into an average rate for
long-term strategy development problems. If the above conditions are fulfilled,
then AB is more appropriate. When modelling only technical systems without
social, behavioural or economic aspects, stationary equilibrium modelling meth-
ods might be useful. However, if there is a need to model social, behavioural and
economic aspects, which is the case in IEPCT (see Sect. 2.3), then complex sys-
tem modelling approaches, such as system dynamics (SD) and agent-based (AB)
simulation are more appropriate. Implementing Evolutionary algorithms into the
optimisation of communal energy supply [17] indicates that the expected calcula-
tion time for good solutions, near the global optimum, is approximately thirteen
weeks. However, in IEPCT, real time simulation and model specification changes
might be required during the planning process with decision makers. This requires
modelling methods that can produce results in certain decision process, which is
usually a daily workshop.

3.3.4 Judgmental-Driven Modelling

To a large extent, these modelling approaches are based on the expertise of people.
This category of models is mainly qualitative or semi-qualitative and the related
methods use heuristics: different cognitive maps such as causal or fuzzy maps for
3.3 3-Domain Modelling: Different Approaches for Different Domains 47

describing interdependence among different system elements or variables, recog-

nition or prediction of patterns, or just weighting interdependency among different
variables using rule-based assessment.
Judgmental methods can vary from easy qualitative, unorganised acts to well
formalised judgmental procedures such as expert elicitation protocols, structured
analogies or judgmental bootstrapping. The progress of judgmental methods over
the last 25 years is reviewed in [18]. In another review, Armstrong [19] propose a
selection tree for choosing an appropriate forecasting method depending on data
availability and character, the problem to be solved and the availability of domain
knowledge. Using a non-well-structured judgmental approach alone for long-term
prediction of neighbouring or distant domains is questionable. Webby [20] indi-
cates, “Thus, the majority of the evidence indicates that judgement performs worse
than objective methods over long forecast horizons.” Braun [21] also agrees that
“Experts outperformed models in shorter-term forecasting, whereas models out-
performed experts in longer-term forecasting.”
In long-term local infrastructure planning, the situation can be quite different.
In certain planning cases, there are some data or domain knowledge available. In
others, there are neither enough data nor domain knowledge available. Therefore, a
combination of different approaches might be more successful. As stated by Bunn
[22], “[…] issue is not that judgement is better or worse than models but that there
are advantages and disadvantages in each approach which are best resolved by
allowing structured interaction of judgement and statistical forecasting methods”.

3.4 Defining Modelling Approaches for Different

Modelling Domains and Use Cases

3.4.1 General

The purpose of this section is to provide a mapping, between previous modelling

approaches and 3-domain Metasystem framework discussed above, taking into
account different study conditions.
Different modelling methods can be implemented for different modelling domains.
Depending on the modelling object complexity, study objective and available
resources and skills, each modelling approach has its own advantages and limits.
For each domain of the two different Use Cases discussed in Sect. 2.4 different
modelling approaches might be required. Whereas judgmental driven model can
be used in any domain the process driven models are more appropriate for targeted
domain. For neighbouring and distant domain domains, data-driven models can
be used if there is sufficient data amount available; otherwise judgmental-driven
models are appropriated.
To define the appropriateness of different modelling approaches for each
domain, pre selection of some methods have been done based on literature review.
One problem in the selection process was that methods had to be selected before
48 3 3-Domain Modelling

they were tested; another was that even if it had been possible to select the meth-
ods on the basis of a priori knowledge or experiences, it would have been impos-
sible to apply a single criterion to all models. It was therefore decided to select
modelling approaches that are widely available and already in use. These model-
ling approaches are then tested in Use Cases and evaluated.

3.4.2 Modelling Approaches for Targeted Domain

3.4.2.1 Selecting the Modelling Methods

In this domain, planning participants are interested for foresight model as well as
for impact and decision model according to the definition in Sect. 1.3.3. However,
depending on use cases Sect. 2.4 the particular needs might differ. No quantita-
tive integrated model for ‘Use Case—I’ was required to be developed at the begin-
ning of planning and modelling. However, in order to have an integrated view of a
city or territory, qualitative approaches such as causal, cognitive maps or a qualita-
tive system dynamic approach can be useful. It is important both to represent the
interaction of different single quantitative sub models, such as models for cooling
a district or models for lighting, and identify important driving forces for all sub
models and their interaction according to influence and dependence. Therefore,
for Use Case—I a qualitative MICMAC approach is selected for targeted domain
modelling and for representation of the interaction of city or territory drivers and
to capture the interdependency of the different sub-models driving forces.
A system dynamic approach is selected for implementation and modelling of
targeted domains for ‘Use Case—II’ because of study needs. However, system
dynamic have some limitations, e.g., in modelling individual spatial depend-
ent fluctuations in a territory. These types of limitations can be overcome using,
e.g., sensitivity analysis to examine the possible trajectories of the system changes
[23]. Aggregated representation of real processes using system dynamics support
to achieve a compromise assessment that avoids any excessive detailed precision
likely to be arbitrary and so uncertain in low data available study conditions. In
such limited conditions and study requirements behavioural modelling approaches
like Agent based simulation was difficult to be implemented.
Detailed descriptions of process-driven modelling methods are presented below.

3.4.2.2 Selected Process Driven Models for Targeted Domain

System Dynamic (SD) Approach to Model the Targeted Domain

in Use Case II—Mexico

A system dynamics approach was developed by Jay W. Forrester during the 1950s
and 1960s [24, 25]. This approach has been used in different areas in control,
3.4 Defining Modelling Approaches for Different Modelling … 49

Connector
Connector
Feedback

Inflow Stock Outflow

Fig. 3.2  Basic components of a stock-flow diagram

organization or in management and decision theory. Forrester provides the ini-

tial implementation of a SD approach to model urban dynamics at the end of the
1960s [26]. SD models usually capture an aggregate view of the system. However,
this approach captures multiple dynamics and interactions of the system. In group
decision making, a causal diagram for the system is usually developed. This dia-
gram is a useful tool in communication. In the next step, causal relations are for-
malized in mathematical terms by transferring causal loop diagrams in stock-flow
diagrams and defining possible feedback relations and time delay effects among
variables and parameters. The picture below shows a simple Stock-Flow example.
The system is described via stocks inflow and outflow. The interactions between
flows and stocks defines the behaviour of the system (Fig. 3.2).
ˆt
Stock(t) = Rate(t)dt
t0

where t is time and Rate is the change rate of the stock and t0 the time at initial
state.
Before implementing the SD model to support decisions, the SD model
is tested and validated using different tests and uncertainty analyses that are
­discussed in Sect. 5.6.8.
The fundamental purpose of a SD model is to assist the understanding and
simulation of the relationships between the behaviour of the system and the
­
­underlying structure over time.
The SD principle and theory is discussed in previous works [27–29]. An updated
description of the SD approach with examples of applications in ­different area have
also been previously provided [30].
50 3 3-Domain Modelling

Influential
scale

1 2

3
4

Dependent scale

Fig. 3.3  Influence-dependence chart

3.4.2.3 Judgment-Driven Modelling Methods

MICMAC Approach to Model the Targeted Domain in Use Case I—Singapore

MICMAC (Matrice d’Impacts Croisés—Multiplication Appliquée à un Classement)

developed by Godet [31] is particularly useful for qualitative structural analysis of
the system to determine the essential variables or factors and for defining the quali-
tative model of the system. Key factors are those variables whose evolution has a
high effect on the system. By contrast to the system grid [32], MICMAC considers
not only the direct relation of factors but also the indirect and potential relation to
identify “hidden” factors. The interactions of factors are presented in a system grid
or influence-dependence chart (Fig. 3.3). MICMAC can be considered a qualitative
system dynamics approach.
The diagram is divided into five sectors [31]:
Sector 1: Input factors. These factors have a high influence and are less depend-
ent. These factors tend to describe the dynamics of a system and the conditions of
the remaining variables or factors. These factors are the first choice for developing
scenarios.
Sector 2: Intermediate factors. These are highly influential and dependent factors.
Any change in these factors will have high flow changes throughout the rest of the
system.
Sector 3: Resultant factors. These are highly dependent factors that have less
influence.
Sector 4: Excluded factors. These are neither influential nor dependent ­variables.
Because of their autonomous character and less connection to the system, they
have little effect on the system; therefore, they can be excluded in the next analysis
steps.
3.4 Defining Modelling Approaches for Different Modelling … 51

Influential scale

Influential scale
Dependent scale Dependent scale
Stable system Unstable system

Fig. 3.4  Allocation of descriptors and model stability

Sector 5: Average factors. These factors cannot be clearly allocated to the remain-
ing sectors. They are not sufficiently influential or dependant.
Additionally, the influence-dependence chart presents the interaction un-/stabil-
ity of system variables. The configuration or allocation of the set of factor-points
in the influence-dependence matrix allows the stability of the entire system to be
determined (Fig. 3.4).
The cloud of points spread along the axis (L shape) indicates that the system
responds to the given impulse of determining variables with a certain degree of
certainty. The system is stable. By contrast, when clouds of variables are spread
along the bisecting line, notably when the points are concentrated in the north-east
frame, the system can be considered as unstable.
Detailed information about MICMAC is discussed by Godet [33–35].

3.4.3 Data Driven Modelling Approaches for Neighbouring

and Distant Domains

3.4.3.1 Selecting the Modelling Methods

In neighbouring domain planning participants have more interest for foresight

model, less interest for impact and no interest for decision model according to
the definition in Sect. 3.2. Foresight models like univariate time series analy-
sis methods and models, S-shaped curves methods or neural networks for pat-
tern recognition can be considered as data driven approaches for neighbouring
or distant domain if there are sufficient data available for both ‘Use Case—I’
and ‘Use Case—II. For long term modelling and extrapolation Makridakis et al.
[36] states that identification and extrapolation of trends of data is i­mportant.
52 3 3-Domain Modelling

Also Makridakis [37] states whereas for short term seasonality, for mid-
term cyclical effect are important, for long term extrapolation trend factors
plays important roll. In the same work it was also argued that “[…] treatments
of complex data sets for extrapolation ARIMA models were indicated to be
appropriate, it optimise data structure and minimise the residue.” On the other
side Makridakis et al. [36] states that less formal approaches are often better.
Therefore several modelling methods have been chosen for implementation and
evaluation. For many modelling approaches like ARIMA models there is a need
to define several model parameters a priory which requires technical and domain
specific knowledge. Therefore automatic model identification, selection and
parameterization of the process can be useful [38] as an expertise is not always
available in cities and territories.
To select data-driven modelling approaches for modelling neighbouring or dis-
tant domains, the following additional criteria where used:
• The method or combination of methods has been successfully implemented
for long-period extrapolation and has shown good performance criteria using
annual data;
• The method is not complex or data intensive;
• It can identify and predict long-term trends and data behaviour automatically in
the case of absent technical competency and when new data are available;
• The method is widely applied and available in existing software or there is at
least enough information to develop script for implementing the method.
It was also important to have methods that are based on a different paradigm, e.g.,
mathematical versus artificial intelligence based, for possible triangulation or
discovering paradoxes.
Reviewing the studies, eight different initial methods have been identified in
Table 3.1 using the criteria discussed before. Certainly, there are other methods

Table 3.1  Preselected methods for extrapolation of driving forces

Methods References Short Description
Linear regression [73] Linear regression
Theta [42, 43] Specific decomposition technique, projection and
combination of the individual components
Automatic ARIMA [45] Automatic ARIMA, Autoregressive-Moving Average
ARIMA (0.2.2) [74] ARIMA model with Stochastic level and stochastic
trend; Holt’s linear trend or robust trend
ANN [54, 73, 75] Feed-forward Artificial Neural Networks with a single
hidden layer
Robust trend [7] Non-parametric version of Holt’s linear model with
median based estimate of trend
S-shaped curve [58, 76] S-shaped curves for diffusion of new technologies
methods
Random walk Random walk with drift (we denot with ‘trend’)
3.4 Defining Modelling Approaches for Different Modelling … 53

that may also be suitable, such as FORSYS [39] or ARARMA [40], which show
good performance for long-term modelling. However, insufficient information
about these methods could be found yet for integrating them in the analysis.
It is also important to analyse and prepare data before implementation of meth-
ods like identifying outliers or possible correlation with other variables. However,
the focus of this book is inductive analysis. Nevertheless, some results of descrip-
tive analysis like identification of distributions or seasonality, trends or stochasticity
are presented in annex A.
A detailed description of each individual method is given in Sect. 3.4.3.2.

3.4.3.2 Selected Data-Driven Modelling Methods

Linear Regression

Linear regression is one of the most commonly used and accepted methods in
energy planning and modelling. A linear regression can capture long-term histori-
cal data trend.
Regression methods are explanatory approaches that establish relationships
between one or more inputs (explanatory variable x) and single outputs (dependent
variable Y). Simple regression establishes relationship between single inputs and
single outputs. If one dependent forecast variable y is related to many explanatory
inputs (x1, x2… xk); this process is called multiple regression.
In a general form, the regression model can be presented as the following:
Yi = f (xi ; β) + εi
where β is the unknown parameter and ε denotes the error.
In simple linear regressions, the model-dependent variables change at a con-
stant rate as the value of the explanatory variable increases or decreases:
Yi = β0 + β1 xi + εi
The objective is to discover β0 (intercept) and β1(slope) so that the line Yi presents
the best fit to the data. The least squares estimation procedure is used to discover
this line of best fit. The least squares principle selects β0 and β1 that minimizes the
sum of squared residuals εi or sum of square error (SSE):
n
 n

SSE = e2i = (Yi − Ŷi )2
i=1 i=1

where ei = (Yi − Ŷi )) is the error for the ith observation and Ŷi is estimated value
of Yi.
However as stated by Armstrong “[…] diagnostic statistics that are commonly
provided with regression analysis may lead to confusion, reduced accuracy, and
overconfidence” [41].
54 3 3-Domain Modelling

Theta Model

The Theta model has been successfully implemented for the numerous series of
the M3 competition which was developed in 1999 [42]. This model proposes a
different approach to decompose the seasonally adjusted series into short- and
long-term components. The basic concept modifies the local curvatures of the time
series by using the Theta (θ) coefficient [42]:
′′ ′′
Xnew (θ ) = θ Xdata
where
′′
Xdata = Xt − 2Xt−1 + Xt−2
at time t.
Xt′′ denotes the second difference of Xt. Smaller θ produce larger degrees of defla-
tion. When θ = 0, the time series is transformed into a linear regression line [42].
Forecasts obtained by the Theta model are similar to simple exponential
smoothing with drift [43]. However, it shows good performance in long range
forecasting. The drift in Theta model is computed as follows:
q 2
X̂n (h) ± σ (h − 1)α 2 + 1
2
where α is the smoothing parameter, q is prediction interval for h period, σ2 is the
variance and X̂n (h) is the point forecast of series Xt.

ARIMA Models

The Autoregressive Integrated Moving Average (ARIMA) was published initially

in the 1970s to apply both time series analyses and forecasting [44]. ARIMA is
a coupled approach of Autoregressive (AR), differencing passes (I) and moving
average (MA) models for analysis of non-stationary series.
Different ARIMA models have been proposed. The general non-seasonal model
is known as ARIMA (p, d, q):
where
p order of the autoregressive part (AR);
d degree of the first difference involving (d); and
q order of the moving average part (MA)
A non-seasonal ARIMA (p, d, q) process is presented by the following equation
[45]:
φ(B)(1 − B)d yt = c + θ(B)εt
where εt is a white noise process with a mean of zero and a variance of σ2, B is
the backshift operator, and φ(B) and θ(B) are polynomials of order p (order of the
autoregressive) and q (order of the moving average), respectively.
3.4 Defining Modelling Approaches for Different Modelling … 55

For certain situations when well-trained users are unable to identify appropriate
extrapolation methods or to handle unusual time series patterns in robust ways, an
automatic forecasting algorithm is useful. Examples of automatic forecasting algo-
rithms are based on ARIMA. An automatic ARIMA forecasting algorithm will be
considered in this book.
To select the appropriate model for a given data set, Akaike’s Information
Criterion is used:
 
AIC = L θ̂ , x̂0 + 2q

where L denotes the likelihood function, q is the number of parameters in θ plus

the number of free states in x0, and θ̂ and x̂0 denote the estimates of θ and x0,
respectively.
The main steps for analysis using auto ARIMA are the following [46]:
• Apply all models that are appropriate for the optimisation of the parameters for
each time series;
• Select the best models according to the AIC;
• Produce point forecasts using the best model (with optimized parameters); and
• Obtain prediction intervals using the bootstrap method by simulating e.g. 5000
future sample paths to find ∝ /2 and 1− ∝ /2 percentiles of the simulated data
at each forecasting horizon.
Detailed information covering the algorithm and a demonstration using both the
M-competition and M3-competition data are available in [46, 47].
ARIMA (0, 2, 2) uses two non-seasonal differences in conjunction with MA
(moving average) terms. It is equivalent of Damped, Holt’s exponential smoothing
method.
Detailed information about ARIMA procedure have been previously published
in [44, 47, 48].

Robust Trend and Random Walk

The robust trend method was initially developed for telecommunication data [49].
“Robust Trend performed best according to all of the relevant error measures used
for all forecast horizons” for the telecommunications data [7]. It is stated that “it is
of interest to discover whether it will perform well on other time series” [7].
At time T, the k step ahead forecast is the following:
XT +k = XT + k µ̂T
where estimator based on observations up to t is as follows
T  
mT  ZT − M T
µ̂T = MT + ψ
T mT
t=1
56 3 3-Domain Modelling

where MT is the median of (Z1 , . . . ZT ) and mT is the median of

(|Z1 − MT |, . . . |ZT − MT |), Zt = Xt − Xt+k for ∀t ∈ {1, . . . , T }.
The response function ψ is of a “three part redescending” type:
ψ(x) = sign(x)max[min(|2x/3|, 1.0, 2 − |x/3|), 0]
“The robustness of the method lies in its treatment of outliers” [7].
More details about the method are available [49].
The estimator for random walk without trend is equal to zero µ̂T = 0. The esti-
mator for random walk with trend which is implemented in R statistic program is
defined as follows
T
1
µ̂T = mean(Zt ) = Zt
T
t=1

Artificial Neural Network (ANN)

Extending the well-known M3-competition with NN3-competion by assessing the

accuracy of more than 60 different forecasting algorithms, “the results of the NN3
suggest that ANN and computational intelligence (CI) methods can perform com-
petitively relative to established statistical methods in time series prediction, but
still cannot outperform them”, and “the efficacy of CI methods relative to statis-
tical methods increases for longer forecasting horizons” [50]. The superiority of
ANN seemed to arise in the later forecast period [51]. A previous study compares
different methods for forecasting with ANNs [51]. This study states that “although
traditional statistical time series methods perform well, many have inherent limi-
tations. First, without expertise it is possible to miss specify the functional form
relating the independent and dependent variables and fail to make the necessary
data trans-formations. Second, outliers can lead to biased estimates of model
parameters. In addition, time series models are often linear and thus may not cap-
ture nonlinear behaviour. Additionally, many traditional statistical methods do not
learn incrementally as new data arrive; instead, they must be re-estimated periodi-
cally.” As advantages, this book argued that “if the underlying model is nonlinear,
then ANN should make better forecasts; this improvement would be increasingly
apparent as the forecast horizon lengthened since the predictions of the nonlinear
model would increasingly depart from those of the linear model over time.”
The main weakness of ANNs is that the process is a black box and not all
results are traceable; a sufficient amount of data is required to train an ANN. One
of the difficult issues in using an ANN is the definition of the network architecture.
Performing a sensitivity analysis to determine the most appropriate network archi-
tecture in 67 % of studies, a previous report concludes that “in general, most
found the use of a single hidden layer effective for the problem being solved”
[52]. Additionally, half of the studies suffer from the validation of the network
architecture or transformation function.
ANN is pattern-recognition method that is a replication of biological neural
network. However, this method only utilizes a limited set of characteristics from
3.4 Defining Modelling Approaches for Different Modelling … 57

Fig. 3.5  A typical FNN with one hidden layer

biology. An ANN can be a feed-forward or a feed-back network depending on

the direction of the information. Figure 3.5 presents a basic Feed forward Neural
Network (FNN). The first layer of a FNN is the input in which the neurons take
information and pass it to the next neurons. In a hidden layer upstream, the infor-
mation is summarized and a non-linear transformation is performed. In the o­ utput
layer, the variable is presented. A FNN is usually trained by gradient descent
­algorithms such as the popular back-propagation.
The general prediction equation for computing a forecast xt (the output) using
selected past observations xt−j1 , . . . , xt−jk as the inputs for the lags (j1 , . . . , jk ) and
h nodes or neurons in a hidden layer for an one hidden-layer ANN can be written
as follows [53]:
 
xt = f0 (ωco + ωho fh (ωch + ωih xt−j ))
h i

where ωch are the weights of the connections between the constant input and the
hidden neurons; ωco are the weights of the direct connection between the c­ onstant
input and the output. ωih and ωho denote the weights for the other connections
between the inputs and the hidden neurons and between the hidden neurons and
the output. The functions fh and f0 denote the activation functions used in the
­hidden layer and in the output, respectively.
Different activation functions exist. The most commonly used functions are
presented in Fig. 3.6.
The implementation of an ANN to forecast exogenous energy model param-
eters is particularly notable because occasionally it is not possible or there is no
need to explain all interrelation among exogenous parameters. For example,
explaining the change in global oil prices by a city planning team using extensive
58 3 3-Domain Modelling

Fig. 3.6  Some commonly used activation functions

process or causal driven models is difficult and is not required. In this situation,
the “black box” syndrome is not a large drawback. In the energy sector, ANNs
have been implemented for long-term load forecasting [54], for monthly energy
demand [55], for forecasting daily urban electric load profiles [56] and for fore-
casting electricity market prices [57].

S-shaped Curve Methods to Forecast Technology Evolution

and Substitution

The growth and diffusion of autonomous systems can be described by a techno-

logical cycle (sometimes called a lifecycle) using S-curves. Different phases are
in the cycle: appearance, growth, maturity and decline. New technologies in the
appearance phase are not marketable, do not have a large diffusion and are not
crucial in market competition. The S-curve concept has been successfully imple-
mented for these questions [58].
Patterns, processes and timescales of the diffusion of new technologies in the
energy sector have been performed in numerous studies [59–63] using S-shaped
diffusion analysis consistent with historical experience and learning phenomena.
Different important conclusions about changes in technology and clean energy
(e.g., “the time constants of technological change are long, decades not years”
or “end-use application drive supply-side transformations”) have been previously
noted [59]. For the technology forecasting of new clean energy, the S-curves
method have been previously implemented to show the level of technological
maturity of hydrogen and fuel cell technologies [64]. A previous analysis of the
acceleration of diffusion in terms of a growth model and determinants of diffusion
rates of small scale technologies, such as mobile telephones [65], compared three
conventional models (Gompertz, Logistic and Bass) to identify the most appropri-
ate model. This analysis concluded that “Empirical results indicate that the most
appropriate model is the Logistic model”.
3.4 Defining Modelling Approaches for Different Modelling … 59

A time-space innovation diffusion model was suggested at the end of the 1970s
as the following equation:
 
∂Ni   km∂ N̄i − Ni
= (a + bNi ) N̄1 − Ni +
∂ti ∂y f(y)i−1,i

When the distance gradient element and space element are constant, then the tradi-
tional logistic time dependent model will be the following:
∂Ni  
= (a + bNi ) N̄1 − Ni
∂ti

3.4.4 Modelling the Distant Domain and Its Impact to Other

Domains

In distant domain planning participants have more interest for foresight model,
less interest for impact and concern for decision model according to the definition
in Sect. 3.2. There is poor knowledge or planning resources to build a process-
driven model for this domain. However, the question of ‘what will happen’ might
still be interesting and necessary for planning. Typically, information about this
domain is collected from different sources or experts. In some situations when suf-
ficiently good quality data is available, data-driven approaches (see Sect. 3.4.3)
can be implemented; otherwise judgmental-driven methods can be used. There is,
however, a need to organise the available information in distant domain and show
the impact on other modelling domains. In next section, a new method Reference
Impact Matrix (RIM) will be discussed for these tasks.

3.4.4.1 Reference Impact Matrix Method (RIM)

A new matrix-based method is suggested in the thesis [5] in order to formalise

and incorporate different knowledge about distant domains. The implemented
categories for describing the situation and upcoming future that can be used for
qualitatively describing the distant domain are taken from the study [66]. Different
categories, such as wild card, weak signals or trends, have been suggested there.
However, these categories have not been organised in a systematic way, and the
ways to incorporate and quantify this information for model-based studies were
not discussed.
The rows of the matrix present different global, transnational, national, local
and model levels. However, other categories like overall or sub branches can be
implementing in RIM for other study needs. The columns present the categories
used by [66] or [67] for other propose. Last rows describe potential typology and
extend of impact particularly to the data-driven models (Table 3.2).
60 3 3-Domain Modelling

Table 3.2  Reference impact matrix

Frame of reference Impact category
Trends-cycles Discontinuities Weak signals Wild cards
(TC) (D) (WS) (WdC)
Global
Transnational
National
Local (City,
territory)
Impacted domain
and model
feature
Type of impact
The extend of
impact

Trends and cycles are general changes that are experienced by everyone over
­periods of several years or decades. Usually, individual organisations or even
nations cannot do much to change them. An example of a trend or cycle is
long-term economic cycles such as Kondratieff waves [68].
“Discontinuities refer to rapid and significant shifts in trajectories without the
aspect of being mostly unanticipated or deeply surprising” [67]. An example of
discontinuities is changes and innovations in technologies.
Weak signals are the first warning signals of possible but not confirmed changes in
future trends or discontinuities. Weak signals are still too incomplete for an accurate
estimation of their impact. An example of a weak signal is global warming [66].
Wild cards or shocks are those surprising events and situations that can happen
but usually have a low probability of doing so—but if they do, their impact is
very high and there is not enough time resources etc. for adequate response. The
changes tend to be fundamental and create new trajectories. Natural disaster or
epidemic are examples of a wild cards.
The category drivers proposed by [66] is not included in RIM because it is
study model specific and will be defined according to individual study model.
The row ‘impacted domain model feature’ shows which domain and entity
or feature of a model is impacted by a particular category, whether it is model
structure, parameter, etc.
There are diverse type of impact, like the changes in the model structure or in
trends, or driving variable behaviour changes. Four types of changes are defined.
Depending on the direction of the causal force in distant domains and trend
­direction of driving forces in other domains, the types of impact are divided into
four sub categories:
• Trend change refers to a change of direction for a certain model variable, such
as energy price.
3.4 Defining Modelling Approaches for Different Modelling … 61

• Transfer changes refer to events that transfer the variable change from one point
in time to another without changing the overall trend in the long range.
• Transient change influences only during a particular event occurs. When an
event is over, the effect does not last any longer.
• Quantum Jump changes occur when the effect of a non-repetitive new event is
very strong or permanent such as in the case of wild cards.
Similar impact types are used by [69] for knowledge assistance and adjustment of
statistical forecasting. Not every trend, discontinuity, etc., will necessarily have an
impact on targeted or neighbouring domains in different study conditions (Fig. 3.7).

Variable
Variable

Time Time
Trend Change
Variable

Variable

Time Time
Transfer Change
Variable

Variable

Time Time
Transient Change
Variable

Variable

Time Time
Quantum jump

Fig. 3.7  Different impact types of causal forces on the data-driven model

62 3 3-Domain Modelling

Table 3.3  Relationship of Type of causal force Driving force direction when

causal forces to trends [70]
Trend is up Trend is down
Growth Up Up
Decay Down Down
Supporting Up Down
Opposing Down Up
Regressing Towards the mean Towards the
mean

To quantify extend of impact on structure or data-driven models, two pos-

sibilities exist. Firstly, rule-based forecasting, as suggested by [70, 71] can be
used. Depending on how the behaviour of the impacted driving force and causal
forces impact direction and character, different possible developments of impacted
­variable are shown in Table 3.3.
A second possibility for incorporating the impact of causal forces (from weak
signals, discontinuities etc.) is using Fuzzy inference. Four parameters, Maximum
Impact, Steady-State Impact, Time to Maximum Impact and Time to Steady-
State Impact, have been used by [72] to evaluate the impact of events. Then,
using Monte Carlo Simulation, the impact and event probability judgments are
combined with the results of the base-case scenario to generate possible future
development. Based on these different developments, the median, 5th and 95th
percentile scenarios are computed to indicate three distinctive developments of a
certain variable. However, this approach is designed for computing the extrapola-
tion of single variable. The interdependency among other model variables using,
e.g., cross-impact or consistency analysis to develop scenarios is not performed in
mentioned study; it is suggested for future research in this reference. This issue is
addressed in this book in Sect. 5.5.2.5.

3.5 Summary of Modelling Approches for Different Use

Cases and Domains

This section presents some modelling-approach candidates for implementation in

different Use Cases and modelling domains. In most cases, combinations, such as
judgmental and data driven approaches, are implemented. However, a particular
modelling approach may play a dominating role for a given Metasystem domain,
such as process-driven modelling for the targeted domain, data-driven modelling
for neighbouring or distant domains, and judgmental-driven modelling for the dis-
tant domain.
For the distant domain, in both Use Cases—I and Use Cases—II, the RIM
method can be implemented. For the neighbouring domain, for both use cases,
eight different data driven and logistic S-curve methods can be implemented.
For the targeted domain, System Dynamics (SD) can be implemented for Use
3.5 Summary of Modelling Approches for Different Use Cases and Domains 63

Table 3.4  Modelling methods, use cases and domains

Use cases Modelling domains
Targeted domain Neighbouring domain Distant
domain
Use case—I MICMAC –Liner regression –RIM
Use case—II System dynamics –Theta
–Automatic ARIMA
–ARIMA (0.2.2)
–ANN
–Random walk with ‘trend’
–Robust trend
–S shaped curve

Case—II and Use Case—I, and the MICMAC qualitative method is more appro-
priate because of the study conditions. An experts’ elicitation is used in all
domains and cases, and therefore, not indicated in Table 3.4.

3.6 3-Domain Modelling in Context of Multi Method

Research

To explain the role and relevance of different modelling approaches in context

of mixed method research, the categories and aspects discussed in Sect. 1.3.4 are
used. The weighing is considered equal for all modelling methods. The ration-
ales using different modelling methods in different domains are expansion and
development. The methods can be used either in sequential or in iterative way.
The methods are combined as a hole therefore it is methodological combination.
Modelling methods used in the same domain concurrently have the triangulation
as a mixed rationale.

References

1. Lorenz T, Jost A (2006) Towards an orientation framework in multi-paradigm modeling. In:

Proceedings of the 24th international conference of the system dynamics society, Nijmegen,
The Netherlands, pp 1–18
2. Klir G, Elias D (2003) Architecture of systems problem solving. Springer, New York
3. Ayyub BM, Klir GJ (2006) Uncertainty modeling and analysis in engineering and the sci-
ences. Taylor & Francis, United Kingdom
4. Ford A, Flynn H (2005) Statistical screening of system dynamics models. System Dynamics
Review 21(4):273–303. doi:10.1002/sdr.322
5. Mirakyan A (2014) PhD-Thesis: Cadre méthodologique pour la gestion de l’incertain en
planification énergétique long terme des villes et territoires. UNI Strabourg, Strasbourg
6. Makridakis S, Hibon M (2000) The M3-competition: results, conclusions and implications.
Int J Forecast 16(4):451–476. doi:10.1016/S0169-2070(00)00057-1
64 3 3-Domain Modelling

7. Fildes R, Hibon M, Makridakis S, Meade N (1998) Generalising about univariate fore-

casting methods: further empirical evidence. Int J Forecast 14(3):339–358. doi:10.1016/
S0169-2070(98)00009-0
8. Ord K, Hibon M, Makridakis S (2000) The M3-competition. Int J Forecast 16(4):433–436.
doi:10.1016/S0169-2070(00)00078-9
9. Mirakyan A, De Guio R (2013) Integrated energy planning in cities and territories: a
review of methods and tools. Renew Sustain Energy Rev 22:289–297. doi:10.1016/
j.rser.2013.01.033
10. Bale CSE, Varga L, Foxon TJ (2015) Energy and complexity: new ways forward. Appl
Energy 138:150–159. doi:10.1016/j.apenergy.2014.10.057
11. Albeverio S, Andrey D, Giordano P, Vancheri A (2008) The dynamics of complex urban
systems. Physica-Verlag Heidelberg and Accademia di Architettura, Mendrisio, Switzerland,
Heidelberg doi:10.1007/978-3-7908-1937-3
12. Walloth C, Gurr JM, Schmidt JA (2014) Understanding complex urban systems: multidisci-
plinary approaches to modeling. Springer, Berlin
13. Borshchev A, Filippov A (2004) From system dynamics and discrete event to practical agent
based modeling: reasons, techniques, tools. In: The 22nd international conference of the sys-
tem dynamics society Oxford, England. doi:citeulike-article-id:5838575
14. Bonabeau E (2002) Agent-based modeling: methods and techniques for simulating human
systems. In: Proceedings of the national academy of sciences. 99(3):7280–7287. doi:10.1073
/pnas.082080899
15. Fagiolo G, Windrum P, Moneta A (2006) Empirical validation of agent-based models: a criti-
cal survey
16. Elert M, Butler A, Chen J, Dovlete C, Konoplev A, Golubenkov A, Sheppard M, Togawa O,
Zeevaert T (1998) Effects of model complexity on uncertainty estimates. J Environ Radioact
42(2–3):255–270. doi:10.1016/S0265-931X(98)00058-7
17. Bouvy C, Lucas K (2007) Multicriterial optimisation of communal energy supply concepts.
Energy Convers Manag 48(11):2827–2835
18. Collopy F, Armstrong JS (1992) Expert opinions about extrapolation and the mystery of the
overlooked discontinuities. Int J Forecast 8(4):575–582. doi:10.1016/0169-2070(92)90067-J
19. Armstrong JS (2005) The forecasting canon: nine generalizations to improve forecast accu-
racy. Int J Appl Forecast 1:29–35
20. Webby R, O’Connor M (1996) Judgemental and statistical time series forecasting: a review
of the literature. Int J Forecast 12(1):91–118. doi:10.1016/0169-2070(95)00644-3
21. Braun PA, Yaniv I (1992) A case study of expert judgment: economists’ probabilities versus
base-rate model forecasts. J Behav Decis Mak 5(3):217–231. doi:10.1002/bdm.3960050306
22. Bunn D, Wright G (1991) Interaction of judgemental and statistical forecasting methods:
issues and analysis. Manage Sci 37(5):501–518. doi:10.1287/mnsc.37.5.501
23. Allen PM (1997) Cities and regions as self-organizing systems: models of complexity.
Gordon and Breach Science Publishers, Philadelphia
24. Forrester JW (1969) Industrial dynamics. MIT Press, Cambridge
25. Forrester JW (1968) Principles of systems: text and workbook chapters 1 through 10. Wright-
Allen Press, Cambridge
26. Forrester JW (1970) Urban dynamics. MIT Press, Cambridge
27. Randers J (1980) Elements of the system dynamics method. MIT Press, Cambridge
28. Roberts N (1983) Introduction to computer simulation: a system dynamics modeling
approach. Productivity Press, New York
29. Richardson GP, Pugh AL (1981) Introduction to system dynamics modeling. Pegasus
Communications, US
30. Sterman J (2000) Business dynamics: systems thinking and modeling for a complex world.
McGraw-Hill, New York
31. Godet M (2001) Creating futures: scenario planning as a strategic management tool.
Économica, Hoboken
References 65

32. von Reibnitz U (1987) Szenarien—Optionen für die Zukunft. McGraw-Hill, New York
33. Godet M (2000) The art of scenarios and strategic planning: tools and pitfalls. Technol
Forecast Soc Chang 65(1):3–22. doi:10.1016/S0040-1625(99)00120-1
34. Godet M (2011) Strategic foresight for corporate and regional development. Unesco, Paris
35. Godet M, Coates JF, Unesco (1994) From anticipation to action: a handbook of strategic pro-
spective. Unesco Publishing, Paris
36. Makridakis S, Wheelwright SC, Hyndman RJ (2008) Forecasting methods and applications.
Wiley, Bengaluru
37. Makridakis SG, Reschke H, Wheelwright SC (1980) Prognosetechniken für Manager.
Gabler, German
38. Hyndman RJ (2011) Forcasting time series using R. MONASH university, Melbourne
39. Lewandowski R (1982) Practitioners’ forum. Sales forecasting by forsys. J Forecast
1(2):205–214. doi:10.1002/for.3980010206
40. Parzen E (1982) ARARMA models for time series analysis and forecasting. J Forecast
1(1):67–82. doi:10.1002/for.3980010108
41. Armstrong JS (2012) Illusions in regression analysis. Int J Forecast 28(3):689–694.
doi:10.1016/j.ijforecast.2012.02.001
42. Assimakopoulos V, Nikolopoulos K (2000) The theta model: a decomposition approach to
forecasting. Int J Forecast 16(4):521–530. doi:10.1016/S0169-2070(00)00066-2
43. Hyndman RJ, Billah B (2003) Unmasking the theta method. Int J Forecast 19(2):287–290.
doi:10.1016/S0169-2070(01)00143-1
44. Box GE, Jenkins G (1970) Time-series analysis, forecasting and control. Holden Day, San
Francisco
45. Hyndman RJ, Khandakar Y (2008) Automatic time series forecasting: the forecast package
for R. J Stat Softw 27(3)
46. Hyndman RJ, Koehler AB, Snyder RD, Grose S (2002) A state space framework for auto-
matic forecasting using exponential smoothing methods. Int J Forecast 18(3):439–454.
doi:10.1016/S0169-2070(01)00110-8
47. Hyndman RJ, Khandakar Y (2008) Automatic time series forecasting: the forecast package
for R. J Stat Softw 27(3):1–22
48. Box GEP, Jenkins GM, Reinsel GC (1994) Time series analysis: forecasting and control.
Prentice Hall, Upper Saddle River
49. Grambsch P, Stahel WA (1990) Forecasting demand for special telephone services: a case
study. Int J Forecast 6(1):53–64. doi:10.1016/0169-2070(90)90097-U
50. Crone SF, Hibon M, Nikolopoulos K (2011) Advances in forecasting with neural networks?
Empirical evidence from the NN3 competition on time series prediction. Int J Forecast
27(3):635–660. doi:10.1016/j.ijforecast.2011.04.001
51. Hill T, O’Connor M, Remus W (1996) Neural network models for time series forecasts.
Manage Sci 42(7):1082–1092. doi:10.2307/2634369
52. Adya M, Collopy F (1998) How effective are neural networks at forecasting and prediction?
A review and evaluation. J Forecast 17(5–6):481–495. doi:10.1002/(sici)1099-131x(1998090)
17:5/6<481:aid-for709>3.0.co;2-q
53. Faraway J, Chatfield C (1998) Time series forecasting with neural networks: a com-
parative study using the air line data. J Roy Stat Soc: Ser C (Appl Stat) 47(2):231–250.
doi:10.1111/1467-9876.00109
54. Al-Saba T, El-Amin I (1999) Artificial neural networks as applied to long-term demand fore-
casting. Artif Intell Eng 13(2):189–197. doi:10.1016/S0954-1810(98)00018-1
55. Abdel-Aal RE (2008) Univariate modeling and forecasting of monthly energy demand
time series using abductive and neural networks. Comput Ind Eng 54(4):903–917.
doi:10.1016/j.cie.2007.10.020
56. Beccali M, Cellura M, Lo Brano V, Marvuglia A (2004) Forecasting daily urban electric
load profiles using artificial neural networks. Energy Convers Manag 45(18–19):2879–2900.
doi:10.1016/j.enconman.2004.01.006
66 3 3-Domain Modelling

57. Pao H-T (2007) Forecasting electricity market pricing using artificial neural networks.
Energy Convers Manag 48(3):907–912. doi:10.1016/j.enconman.2006.08.016
58. Kucharavy D, De Guio R (2011) Application of S-shaped curves. Procedia Engineering
9:559–572. doi:10.1016/j.proeng.2011.03.142
59. Wilson C, Grubler A (2011) Lessons from the history of technological change for clean
energy scenarios and policies. Nat Resour Forum 35(3):165–184. doi:10.1111/j.1477-
8947.2011.01386.x
60. Grübler A, Nakićenović N, Victor DG (1999) Dynamics of energy technologies and global
change. Energy Policy 27(5):247–280. doi:10.1016/S0301-4215(98)00067-6
61. Grubler A (2012) Energy transitions research: insights and cautionary tales. Energy Policy
50:8–16. doi:10.1016/j.enpol.2012.02.070
62. Wilson C, Grubler A, Gallagher KS, Nemet GF (2012) Marginalization of end-use tech-
nologies in energy innovation for climate protection. Nat Clim Change 2 (11):780–788.
doi:http://www.nature.com/nclimate/journal/v2/n11/abs/nclimate1576.html#supplement
ary-information
63. Wilson C, Grubler A, Bauer N, Krey V, Riahi K (2013) Future capacity growth of energy tech-
nologies: are scenarios consistent with historical evidence? Clim Change 118(2):381–395.
doi:10.1007/s10584-012-0618-y
64. Chen Y-H, Chen C-Y, Lee S-C (2010) Technology forecasting of new clean energy: the
example of hydrogen energy and fuel cell. Afr J Bus Manage 4(7):1372–1380
65. Chu W-L, Wu F-S, Kao K-S, Yen DC (2009) Diffusion of mobile telephony: an empirical
study in Taiwan. Telecommun Policy 33(9):506–520. doi:10.1016/j.telpol.2009.07.003
66. Saritas O, Smith JE (2011) The big picture—trends, drivers, wild cards, discontinuities and
weak signals. Futures 43(3):292–312. doi:10.1016/j.futures.2010.11.007
67. Saritas O, Nugroho Y (2012) Mapping issues and envisaging futures: an evolutionary scenario
approach. Technol Forecast Soc Chang 79(3):509–529. doi:10.1016/j.techfore.2011.09.005
68. Solomou S (1990) Phases of economic growth 1850–1973: Kondratieff Waves and Kuznets
Swings. Cambridge University Press, Cambridge
69. Jae Kyu L, Sang Bong O, Jung Cheol S (1990) UNIK-FCST: knowledge-assisted adjustment
of statistical forecasts. Expert Syst Appl 1(1):39–49. doi:10.1016/0957-4174(90)90067-5
70. Scott Armstrong J, Collopy F (1993) Causal forces: structuring knowledge for time-series
extrapolation. J Forecast 12(2):103–115. doi:10.1002/for.3980120205
71. Adya M, Collopy F, Armstrong JS, Kennedy M (2001) Automatic identification of time
series features for rule-based forecasting. Int J Forecast 17(2):143–157. doi:10.1016/
S0169-2070(01)00079-6
72. Agami NME, Omran AMA, Saleh MM, El-Shishiny HEE-D (2008) An enhanced
approach for trend impact analysis. Technol Forecast Soc Chang 75(9):1439–1450.
doi:10.1016/j.techfore.2008.03.006
73. Torgo L (2011) Data mining with R: learning with case studies. Chapman & Hall/CRC,
United Kingdom
74. Meade N (2000) Evidence for the selection of forecasting methods. J Forecast 19(6):515–535.
doi:10.1002/1099-131x(200011)19:6<515:aid-for754>3.0.co;2-7
75. Balkin SD, Ord JK (2000) Automatic neural network modeling for univariate time series. Int
J Forecast 16(4):509–515. doi:10.1016/S0169-2070(00)00072-8
76. Schilling MA, Esmundo M (2009) Technology S-curves in renewable energy alternatives:
analysis and implications for industry and government. Energy Policy 37(5):1767–1781.
doi:10.1016/j.enpol.2009.01.004
Chapter 4
Conceptual Basis of Uncertainty in IEPCT

The objectives of this chapter are

• Presentation of the need for incorporating uncertainty analysis in IEPCT
• Presentation of a conceptual basis of uncertainty and a typology of uncertainty
in context of IEPCT
• Showing the current state of IEPCT studies coping with uncertainties explicitly

4.1 Why Be Explicit About Uncertainty in IEPCT?

To a large extent, the discussion in this chapter is based on the uncertainty ­concept
discussed in Sect. 1.3.5 and proposed by Mirakyan and De Guio [1]. Different
types about uncertainty in the context of integrated energy planning are presented.
Detailed information about different uncertainty types and a literature review is
provided in [1].
Different reasons have been presented and discussed regarding the need for
uncertainty analysis in infrastructure planning in [1]. Flyvbjerg et al. [2] argue, “In
the public sector, the establishment of a credible risk management plan should be
a part of the documentation required before any decision is taken on whether to go
ahead with a project or not.” The authors present different reasons for prediction
of failures such as discontinuous behaviour and the influence of complementary
factors or unexpected changes of exogenous factors [2]. The need for uncertainty
analysis is also indicated by [3–6].

© Springer International Publishing Switzerland 2015 67

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_4
68 4 Conceptual Basis of Uncertainty in IEPCT

4.2 Typology of Uncertainty

Based on the literature review about typology of uncertainty in public decision

context, on decision theory and on conceptualisation of uncertainty in different
theories [1] provides conceptual basis of uncertainty in context of IEPCT.
In general uncertainty can be defined as any departure of inaccessible notion of
determinism. Uncertainty can be determined using different types according to
• the nature like episteme (limited knowledge) and aleatoric (variability)
uncertainty.
• the location in the model like model inputs, model structure or model output.
• the linguistic aspects like ambiguity or vagueness
• the decision making activities: like identification of objectives and solutions
or selection of them
• the procedure like communication or planning project resources limitation.
Linking the uncertainty typology with modelling steps discussed in Sect. 2.5
­following table Table 4.1 is proposed by [1].

Table 4.1  Uncertainty types in modelling and planning process ([1] used with permission of
ELSEVIER)
Uncertainty typology Planning steps
Planning phase I Planning phase II Planning phase III
and IV
Modelling steps
Mental Expressed Conceptual Formal Computer Applied
model model model model model model
(verbal…)
Epistemic, Context and framing x x
reducible or Model inputs and x x! x
knowledge parameters
uncertainty
Model structure x
Model technical x
Model output x
Linguistic Vagueness x!
uncertainty Ambiguity x
Under specificity x x
Decision- Goals/objectives x x
making Actions/strategies x x
uncertainty
Assessment criteria
performance measure
Procedural Communication x x x x x
uncertainty
Available time, x x x x x x
recourses
Key x: uncertainty type appears and is present notably in the related planning and modelling step,
if this type of uncertainty is not addressed at this planning step it will remain in all planning
phases “!” This uncertainty type is addressed in this step reminded in IEPCT literature yet
4.2 Typology of Uncertainty 69

4.2.1 Linguistic Uncertainty

Linguistic uncertainty is defined “as uncertainty that arises because our natural
language is vague, ambiguous, and because the precise meaning of words can
change over time”. Linguistic uncertainty are presented using three sub-­categories.
Vagueness uncertainty arises because of limitations in natural and scientific
­language that cannot precisely describe quantity in some cases (i.e., probability
of some unexpected event or change of lifestyle) [7]. “Ambiguity arises because
some words have more than one meaning and it is not clear which meaning is
intended” [7]. Under-specificity can also arise as a result of epistemic uncertainty
[7] which refers to generalisation or specification of some information.
Uncertainties related to model framing, model inputs or structure etc. are
referred to an overall category denoted epistemic or knowledge uncertainty.

4.2.2 Epistemic Uncertainty

Context and framing uncertainty refers uncertainty in planning context framing,

boundary conditions setting or stakeholder perception of reality and uncertainty.
Definition of planning and modelling context is usually done in planning phase I.
Detailed discussion of planning context uncertainty and demonstration in a use case
is provided by [8].
Model inputs and parameters. Input data can be divided into two groups
(a) external driving forces and (b) model system data. External driving forces are
inputs that are exogenous and describe surroundings. Model inputs uncertainty
are associated with values of model parameters and external driving forces. The
uncertainty arises because of measurement error or because of misclassification
and estimation of parameters or external driving forces through no representative
samples caused by time, space, knowledge or financial ­limitations [1, 7].
Model structure uncertainty is because of incomplete understanding of real sys-
tem processes like physical or economic process and because of system theoretical
point of view when simplifying complex dynamic system for model representation
[1]. The uncertainty in model structure arise through the approximation of a func-
tional form of dynamic interrelation among several parameters and variables of the
model, or through (dis-) aggregation of different system components in time and
space.
Model technical uncertainties arise through software and hardware errors.
These errors arise from software or hardware bags [1].
Model output uncertainty is the accumulation of all other uncertainties (data,
parameters, structure etc.) [1].
70 4 Conceptual Basis of Uncertainty in IEPCT

4.2.3 Variability Uncertainty

Variability uncertainty is due to inherent variability manifested in natural and

human systems [1]. This type of uncertainty can be technological i.e., technology
diffusion patterns or natural changes i.e. temperature or wind.

4.2.4 Decision Making Uncertainty

Decision uncertainty arises whenever there is ambiguity or controversy about how

to define, quantify, compare or select objectives or initiatives. The assessment of
importance of objectives, which is highly dependent on risk perception of the deci-
sion maker, is another uncertainty type that can be viewed very differently by dif-
ferent planning participants [1, 7].
Decision uncertainty rise also because of traps particularly arising in planning
phase III when strategies or planning initiatives have to be selected for sustain-
able development. The four ‘survival’ dilemmas benefit-risk, temporal, spatial, and
social or their combination can lead into traps and have significant impact to the
decision. In a simple case “[…] small benefits are overweighed compared to future
large expected losses. The combination of different traps lead to an overweighting
of ‘benefits, now, here and for me’ compared to future risks affecting a ‘later for
wider area and for collective of people’ [9].

4.2.5 Procedural Uncertainty

Procedural uncertainty refers to available resources or time and communication.

Determining new information or data might require additional, but unplanned
time or resources. However, it is not clear a priori whether investing more
resources and time will increase the quality of information for modelling or plan-
ning. Uncertainty in communication arise when the information is not properly
explained and documented to different planning participants.
Uncertainty arise also because of so called ‘cognitive anomalies’:
­circumstances in which individuals exhibit surprising departures from rationality
[10] like Availability: Responses rely too heavily on readily retrieved i­nformation,
and too little on background information or Endowment: Current status and
­history are favoured relative to alternatives not experienced. Uncertainty related to
­cognitive biases which are present in planning phase I and particularly in phase III
are not discussed in this book.
4.2 Typology of Uncertainty 71

4.2.6 Levels of Uncertainty

Different levels of uncertainty proposed in and discussed in Sect. 1.3.6. There are
• Level 0: Decision in deterministic environments
• Level 1: Decision in risk laden environments
• Level 2: Decision in uncertain environment
– Level 2.1: Decision in environments described by fuzzy probabilities
– Level 2.2: Decision in environment with environments described by a belief
and plausibility function
– Level 2.3: Decision in environments described by a possibility membership
function
• Level 3: Decision in ignorant environment
These levels of uncertain environments cannot be ranked according to their confi-
dence level coherently because the conceptualisation of uncertainty in the environ-
ments is formalized differently by different theories [1].

4.3 Incorporating Uncertainty in Current IEPCT Studies

Reviewing the literature about implementation of uncertainty in context IEPCT [1]

summarize that
• IEPCT involves different typologies of uncertainty in different planning and
modelling steps.
• Most of IEPCT studies are deterministic. Models are developed based on aver-
age or central tendency of model parameters or variables.
• There is a need to identify all aspects of uncertainty, quantify their impact
on energy planning and modelling output systematically, and address them
­explicitly during the planning and modelling process as an integrated part and
not in the end when the plan or model is ready.
• Most of the reviewed studies used scenario analysis techniques and expert
­judgment to generate a single scenario that is used later, for implementation of
different initiatives and for impact assessment.

4.4 Conclusion

Different uncertainty types exist in integrated energy planning. These types of

uncertainty are framed in a consistent typology of uncertainties in the context of
IEPCT. The discussed framework will be implemented in chap. 5 for the evalu-
ation of different uncertainty analysis methods and techniques. This concept is
also used for the development of a general methodological framework, which
72 4 Conceptual Basis of Uncertainty in IEPCT

integrates different methods for dealing with different uncertainty types d­ uring
the planning process in a holistic and complementary way based on different
paradigms.

References

1. Mirakyan A, De Guio R (2015) Modelling and uncertainties in integrated energy planning.

Renew Sustain Energy Rev 46:62–69. doi:10.1016/j.rser.2015.02.028
2. Flyvbjerg B, Bruzelius N, Rothengatter W (2003) Megaprojects and risk: an anatomy of
ambition. Cambridge University Press, Cambridge
3. Hodge B-MS, Huang S, Siirola JD, Pekny JF, Reklaitis GV (2011) A multi-paradigm
modeling framework for energy systems simulation and analysis. Comput Chem Eng
35(9):1725–1737
4. Hirst E, Schweitzer M (1989) Uncertainty: a critical element of integrated resource planning.
Elect J 2(6):16–27. doi:10.1016/1040-6190(89)90022-5
5. Loulou R, Kanudia A (1999) The kyoto protocol, inter-provincial cooperation, and energy
trading: a systems analysis with integrated MARKAL models. Energy Stud Rev 9(1):1–23
6. Klauer B, Brown JD (2004) Conceptualising imperfect knowledge in public decision
making: ignorance, uncertainty, error and ‘risk situations’. Environ Res Eng Manage
27(1):124–128
7. Ascough Ii JC, Maier HR, Ravalico JK, Strudley MW (2008) Future research challenges for
incorporation of uncertainty in environmental and ecological decision-making. Ecol Model
219(3–4):383–399. doi:10.1016/j.ecolmodel.2008.07.015
8. Mirakyan A, De Guio R (2014) A methodology in innovative support of the integrated
energy planning preparation and orientation phase. Energy 78:916–927. doi:10.1016/
j.energy.2014.10.089
9. Vlek C, Keren G (1992) Behavioral decision theory and environmental risk management:
assessment and resolution of four ‘survival’ dilemmas. Acta Psychol 80(1–3):249–278.
doi:10.1016/0001-6918(92)90050-N
10. McFadden D (1999) Rationality for economists? J Risk Uncertainty 19(1–3):73–105. doi:
10.1023/A:1007863007855
Chapter 5
Multi-method Approaches for Uncertainty
Analysis

The objectives of this chapter are

• Demonstrating the principles of design of uncertainty analysis methodologies
• Identifying and formulating analytical sophistication degrees for uncertainty
analysis in the context of IEPCT
• Suggesting and implementing quality factors for evaluation of uncertainty analysis
methods and methodologies, thus creating an extensive review of different methods
• Presenting two different multi-method approaches for uncertainty analysis
based on probabilistic and fuzzy set theory
• Providing a review of the methods and methodologies for uncertainty analysis

5.1 Introduction

Different themes and aspects need to be considered when designing a multi-

method approach [1]. Figure 5.1 shows all of the main themes allocated in
different sections and chapters that are considered when multi-method approaches
for uncertainty analysis need to be designed.
The uncertainty typologies presented in Chap. 4 lead us to highlight some voids
in the present praxis of uncertainty analysis in IEPCT. These voids provide us with
the motivation for exploring available alternative methods that could be used to fill
them. Because we would like to use them in IEPCT, it is first necessary to examine
the review evaluation criteria of these methods and the degree of analytical
sophistication.

© Springer International Publishing Switzerland 2015 73

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_5
74 5 Multi-method Approaches for Uncertainty Analysis

Quality factors
Analytical Mixed method
sophistication degree; Section 5.3 research, chapter 1,
Section 5.2 Section 1.3.4

Design of multi
Uncertainty types, method approach Individual methods
Chapter 4, section 4.2 for uncertainty Section 5.6
analysis (UA)

Modelling approaches and

IEPCT, specific conditions,
use cases modelling domains
chapter 2, section 2.4 Chapter 3

Fig. 5.1  Themes and aspects influencing the design of multi-method approaches for uncertainty
analysis

5.1.1 IEP in Cities and Territories, Specific Conditions

The planning and modelling process in cities and territories, as mentioned in

Sect. 2.2, is an iterative process. However, there is a main stream, sequences
of planning and modelling tasks that can be divided into four planning phases.
Different types of uncertainty are allocated in different planning and modelling
steps; therefore, incorporation of uncertainty must be performed taking into account
modelling and planning phases and steps as well as the Use Cases discussed in
Sect. 2.4. Two different ‘Use Cases’ are encountered separately in Chap. 6.

5.2 Analysis Sophistication Degrees

5.2.1 Introduction

Different degrees of analytical sophistication in the treatment of uncertainties

in risk analysis and the possibility of transfers of experience across the field of
application are proposed by Paté-Cornel [2]. Four different degrees of uncertainty
analysis, 0 to 4, were proposed. A new degree of analytical sophistication of
analysis based on experience and knowledge about fuzzy set theory is additionally
proposed in the thesis [1].
This section briefly presents the different degrees (the term “level” used by
Paté-Cornel is replaced by the term “degree” in the book to keep the terminology
consistent with the rest of the terminologies in this book) in the treatment of
uncertainty by [2].
5.2 Analysis Sophistication Degrees 75

Degree 0: hazard detection and failure modes identification

Hazard detection and failure identification simply involve ways of identifying
how a system can fail by means of qualitative judgment, experiments or past
experience. This degree of analysis may be appropriate for supporting zero-risk
policies or in decision situations when costs are low for known unique solution.
However, it is not possible to rank risk mitigation measures according to their
cost-effectiveness [2], which is required in IEP in cities and territories.

Degree 1: ‘Worst-case’ approach

“It is based on the worst-case assumptions and yields, in theory, the maximum
loss. It is reasonable if the worst loss ‘what do I risk?’ is sufficient to support the
decision” [2]. However, in the development of a city or territory strategy, it is often
possible to imagine numerous unlikely situations that worsen the results, thereby
making the trade-off between the different solutions difficult. Degree 1 does not
consider any notion of probability.

Degree 2: ‘Quasi-worst cases’ and plausible upper bounds

More practical than the ‘worst case’ is the ‘quasi-worst case’ if either there is some
uncertainty as to what the worst case might be or the worst case is so unlikely that
it is meaningless. This degree is an appropriate cut-off point for policy decisions
if residual risk is not known. However, if the bound is large, ‘plausible upper
bounds’ does not allow meaningful comparison of the risks of cost benefit analysis
for different strategies under budget constraints [2]. Like degree 1, degree 2 does
not consider any uncertainty structure using e.g. probability.

Degree 3: ‘Best estimate’ and central values

Degree 3 involves a ‘best estimate’ on the central value (e.g., the mean or median)
of the model variable. “The mean has the theoretical advantage of being the
best basis for economic optimisation and the disadvantage of being sensitive to
extreme values. Generally speaking, the advantage of central values is to provide a
reasonable balance to plausible upper bounds. Their disadvantage is that the risk is
still characterised by a single point estimate” [2].

Degree 4: Probabilistic Risk Assessment (PRA), single risk distribution

“In its simplest form, PRA can be performed to obtain a distribution of the
probabilities of different system states (for example on a continuous spectrum)
based on best estimates of the models and parameter values” [2]. Several types
of uncertainty, such as epistemic and aleatory uncertainty, are aggregated into one
distribution curve. It is not possible to extract their share of contribution.

Degree 4a: Fuzzy scenario based assessment (FSU)

PRA makes the assumption that enough data are available for defining relevant
statistical measures. However, the review in [1] shows that there is less probability
information available or probability can be only defined roughly. Scenario
approach which is used in such a situation based above all on qualitative
76 5 Multi-method Approaches for Uncertainty Analysis

information, on the experiences of experts which are mostly not precise

formulated. An alternative to PRA that uses fuzzy proposition, which is closely
linked with possibility theory is proposed in [1]. FSU based assessment can
combine different paradigms namely fuzzy set and probability theory. The fuzzy
framework is more appropriate in such a situations where less and not precise
data are available. Fuzzy proposition makes possible to cope with this limitation
in scenario approach. Similarly to degree 4, degree 4a provides analysis without
separating the different uncertainty types in the results.

Degree 5: Probabilistic risk analysis, multiple risk curves

This is the highest degree of sophistication for uncertainty treatment in these
­categories. Different types of uncertainty are analysed separately: “Unlike degree 4,
the uncertainties are kept separated in the analysis and the result is a family of risk
curves that provides, for each value of the potential losses, a discretisation of the
probability distribution of the future frequency of expedience of this loss value” [2].

5.2.2 Appropriate Analytical Degrees in IEPCT Context

As was already mentioned, IEPCT has budget constraints and, among other
criteria, planning costs are important. Paté-Cornel concludes “Again, full
treatment of uncertainties may or may not be needed depending on the decision
to be made. However, a gradual move from level 2 to level 4 is desirable when
cost-effectiveness is an issue.” The degree 3 analytical sophistication is the most
implemented degree in energy planning studies, as seen in the review in [3, 4].
Degree 3 will be used as a reference for comparing the discussed approaches in
the Chap. 7. Distinct representations of uncertainty types, which are required when
implementing ‘analytical sophistication degree 5’, are stated to be important in
[5, 6]. However, as indicated by Hofer [7], “Separation of uncertainties is costly
because there is effort involved in consistent separation
• expert judgement elicitation has to account for separation
• propagation through the model needs to happen separately
• presentation has to cater for two uncertainty dimensions
• sensitivity measures are of interest in both dimensions and depend on situations
of practical relevance.”
Because of the limited expertise for the different types of uncertainty for given
case studies, degree 5 is not considered in the next chapters.
The final remaining candidates for the analysis are degree 4, 4a and 3 as
reference. The degree 4 single risk distribution is the one candidate. Another
candidate is degree 4a, which includes also degree 2 but provides additional
opportunity to present not only possible bounds, but many different scenarios also
information about consistency and, if required, the rough probability of scenarios.
5.3 Quality Factors of Methods for Uncertainty Analysis 77

5.3 Quality Factors of Methods for Uncertainty Analysis

The overall quality of the methods used for uncertainty analysis depends not only on
addressing all uncertainty types discussed in Chap. 4 but also on other overall quality
factors presented in Sect. 2.6 in Chap. 2. These quality factors, which are specified
below for uncertainty analysis methods, can provide a framework for reviewing,
evaluating available methods and for developing methods or methodology.

5.3.1 Technical Quality Factors

• Completeness of uncertainty types addressed. The completeness of incorpo-

ration of uncertainty types. Due to previous reviews, analysing and assessing all
relevant uncertainty types according to the planning and modelling process, as
discussed in Sect. 4.2 in Chap. 4, are required.
• Incorporating quantitative and qualitative information: Different informa-
tion can be available in IEP for cities and territories. Some of this information
is quantitative, such as statistical data; other information is qualitative, such as
expert judgment. Both sources of information are important and valuable for
long-term IEPCT because usually there is little information available. This factor
refers to the capability to incorporate and integrate both information.
• Automatism refers to whether the method allows planning participants to
generate new results automatically when the situation changes, new information
is available or some modification is required. It is useful for reducing the
modelling and planning time and the need for subjective judgment in cases
where experts are not available. It is useful that the method automatically adapt
the model to new information conditions with less subjective judgment.
• Mixing synergy refers to the additional quality from using multiple methods
instead of one method in the sense of mixed method research rationales,
combined use, etc. (see Sect. 1.3.4, in Chap. 1).

5.3.2 Organisational Capability

• Required resources and time: The computational time for using methods
has been stated to be an important issue in the last decades. However, these
aspects have become less significant because of increasing computer power and
decreasing associated expenses. More important are the time and money required
for collecting information from different sources, including experts in order
to use the given method completely. For example, there is a need to have not
only upper and lower bounds of a certain value but also the entire probability
distribution to implement Monte Carlo analysis fully. This quality factor can be
evaluated using qualitative ordinary measures like low, middle or high.
78 5 Multi-method Approaches for Uncertainty Analysis

• Another quality factor for evaluating information requirements, is presented

using 5 different levels of information intensity (Int.):
– Int. 01: No cardinal data are required; linguistic description, such as increase,
small etc.
– Int. 02: Linguistic terms with a fuzzy membership function
– Int. 03: Central value of data, most expected
– Int. 04: The interval measures for data, up and down boundaries
– Int. 05: Probability distribution of data and relevant measures
The time and resources increase accordingly when the method is data intensive.
• Expertise and skills: Diverse expertise or skills are required for implementing
certain methods. Five different degrees of expertise and skills can be
distinguished according to the level of analytical sophistication discussed in the
previous section. The higher the degree of analytical sophistication, the more
expertise and skill are required. Expertise and skills refer here to the technical,
analytical knowledge necessary for using certain methods and interpreting the
results independently from the domain knowledge (e.g., knowledge about energy
production systems or economic development in the city). This quality factor can
be evaluated using qualitative ordinary measures like low, middle or high.
• Flexibility refers mainly to the capability of the method to be implemented in
different planning and modelling phases. This quality factor refers also to the
appropriateness of a method or methodology for two different specific planning
condition ‘Use Cases’ discussed in Sect. 2.4 or modelling domains in Sect. 3.2.
• Legitimacy refers to the quality of uncertainty analysis methods used to fit
formal requirements, such as regulations and directives relating to certain
restrictions, or procedures when making uncertainty analysis in cities or
territories planning and modelling (Table 5.1).

5.3.3 Satisfaction by Planning Participants

• Satisfaction working with methods mainly depends on the level of familiarity

the planner or analyst has with the methods they use. When the methods are
completely new for them, it requires time and efforts to learn, accept and
implement them. The level of satisfaction is defined by asking the user in the
implementation in Use Cases about satisfaction and familiarity with a method.
Two levels are defined, high and low.
• Satisfaction of decision maker with methods includes many other q­uality
factors, such as required time and resources as well as flexibility. However, one
particular aspect which influence this category is ways of ­communicating results
to decision maker (DM). As stated in [8, 9], the communication of results is very
important in policy decision making. Using the idea of “progressive disclosure of
information” and considering uncertainty from the decision maker perspective,
Kloprogge provides in [9] criteria for uncertainty communication. For example,
5.3 Quality Factors of Methods for Uncertainty Analysis 79

Table 5.1  Quality factors of the methods for uncertainty analysis

Overall quality factors Quality factors Measure of quality and
comments
Technical quality factor Completeness of uncertainty Completeness of uncertainty
types addressed types addressed
Incorporating qualitative Qualitative, quantitative or mix
and quantitative information
Automatism Level of automatic adaptation
of model parameter
Mixing synergy Different aspects of mixed
method research
Organisational capability Required time and resources Different ordinal level: low,
middle and high
Required information Information intensity levels
Required expertise and skills Level of expertise or skills
Flexibility Level of flexibility to
implement the method in
different planning conditions,
in defined use cases in book
Legitimacy Formal relevancy for certain
planning conditions
Satisfaction by planning Satisfaction working Level of familiarity with
participants with method method
Satisfaction of decision maker Different aspects by
with methods communicating and evaluated
results

uncertainty information should be consistent across different issues, reports, etc.,

and uncertainty should meet the information needs of the target audience, which
is context dependent and needs to be customised. This aspect, similar to previous
ones, will be assessed during the implementation of approaches in Use Cases
discussed in implementation in Chap. 6.

5.4 Methods and Methodologies for Uncertainty

Assessment: A Review

A detailed review of all of the methods and references of the reviewed papers is
provided in Sect. 5.6. In this section, we provide a summary of the results.

5.4.1 Evaluation Criteria

The most commonly used methods for uncertainty analysis in environmental

modelling are presented by [10, 11]. The methods are described, and applications for
80 5 Multi-method Approaches for Uncertainty Analysis

different types of uncertainty are provided. Therefore, only the methods that are not
discussed in these papers or the methods that are used for designing the multi-method
approach for uncertainty analysis are discussed in this book, in Sect. 5.6.
The items that have been used for the review consist of the attributes suggested
in [11] for presenting and describing uncertainty, the quality factors of the
previous section and additional attributes about the application area. The last part
is added to show for which Use Case and for which modelling domain the method
application is most appropriate. The following is a list of used item categories:
• Brief description of the method
• What type of uncertainty does or can this method address?
• Potential rationales in the context of mixed method research
• Method performance: How is the method rated according to the quality factors
defined above?
• References to handbooks, user guides, example case studies and web sites are
also provided for each method, for future reading.

5.4.2 List of the Reviewed Methods and Methodologies

The most widely available and commonly used methods are selected for review.
Some of them have been already implemented in the area of energy planning and
modelling see Chap. 4.
1. Correlations, copula
2. Expert Elicitation
3. Fuzzy inference
4. Innovative multimethod approach
5. Inverse modelling (predictive uncertainty)
6. Interval prediction
7. Monte Carlo Analysis
8. Multiple model simulation
9. NUSAP (Numeral Unit Spread Assessment Pedigree)
10. PRIMA (Pluralistic Framework of Integrated uncertainty Management and
risk Analysis)
11. Scenario Analysis and Fuzzy clustering
12. Sensitivity Analysis
13. Tests of complex models

5.4.3 Summary of Evaluation Results of Reviewed Methods

A review of each method according to the concepts discussed in the above sections
is presented in Sect. 5.6. The results of the evaluation can be summarised as
follows:
5.4 Methods and Methodologies for … 81

• There are different methods for coping with different types of uncertainty.
However, none of them can address all the types of uncertainty.
• The methods for supporting a detailed analysis of some uncertainty types, such
as uncertainty in ignorance situations, could not be found.
• The reviewed methods show significant differences according to the quality
criteria defined in Sect. 5.3.
• There are comprehensive methodologies such as NUSAP or PRIME that
can be used in combination with other methods. However, the methodology
discussions are only focused on environmental or process-driven models and
do not cover other modelling domains. Moreover, the combination of different
individual methods depending on the Use Case and planning or modelling
procedure is not specified yet and uncertainty typology used differs from that
what is presented here.

5.5 Multi Method Approaches for Uncertainty Analysis

5.5.1 Introduction

Two uncertainty analysis approaches are presented in this section, proposed by

[1]; Fuzzy-Scenario based Uncertainty Analysis approach (FSUA) according
to degree of analytical sophistication 4a (Sect. 5.2) and a Probabilistic, Random
Sampling based Uncertainty Analysis approach (PRSUA) according to degree of
analytical sophistication 4. Both approaches are multi-method approaches that
include different methods that are linked to each other using different aspects of
mixed method. The allocation, mixing rationales of different methods and the
planning and modelling procedure are discussed accordingly. In the next sections,
the process of both approaches is presented, and required methods for supporting
them are presented. A detailed description and evaluation of each individual
method is provided in Sect. 5.6.

5.5.2 Fuzzy Scenario Based Uncertainty Analysis

for Use Case-I

5.5.2.1 Analysis Procedure and Steps, Functional View

In Use Case-I there is no integrated, quantitative process (economic or technical)

driven model required to be developed initially. Several models might be
constructed independently for different sectors, such as cooling system or
transportation, but they are not in an integrated quantitative model and might be
developed later outside of integrated project planning process. However, for such
a situation, it can be still important to analyse overall uncertainty not only for
82 5 Multi-method Approaches for Uncertainty Analysis

individual models but for the context of the whole city or territory considering
interdependency among several sub-models at least in a qualitative way.
For this situation, when there is not enough quantitative information and most
of the information are qualitative with non-precise values, Fuzzy proposition and
Possibility theory-based methods are more appropriate, as discussed in Sect. 1.3.7
Chap. 1 and as concluded by [12], “[…] problems involving a complicated
weave of technical, social, political, and economic factors and decisions; and
problems with incomplete, ill-defined, and inconsistent information, where
conditional probabilities cannot be supplied or rationally formulated, perhaps
are candidates for fuzzy logic applications”. As discussed in Chap. 2 Sect. 2.3
energy system in cities and territories is sociotechnical system and decision has
to be done in uncertain environment where no probability or only some fuzzy
probability information is available. Therefore, FSUA approach is designed using
the methods of fuzzy proposition to use qualitative information efficiently and
decrease linguistic uncertainty. Particularly, fuzzy clustering method and fuzzy
inference is implemented to incorporate non prices information and probabilities.
The overall uncertainty analysis, according to the analytical sophistication 4a level
Sect. 5.2, has different functional steps, which require different individual methods.
The functional steps and involved individual methods are defined according to
IEPCT planning and the modelling process discussed in Sect. 2.2. The uncertainty
analysis, combined use of different methods and stream of information have been
defined according to the mixed method research aspects discussed in Chap. 1. The
functional steps and the functional view of the analysis are presented in the Fig. 5.2.
Methods for supporting decision or model context uncertainty in the initial planning
phase are proposed and discussed in [13]. The functional step ‘model technical
uncertainty’ is not presented in the picture because its functional position is not
fixed. It can be implemented after the 3rd or 4th function. The main restriction
for all functional steps is the availability of data, information, time resources and
expertise. Each box performs certain functions, have input and output information
and require methods and people to perform the function under given constraints.

5.5.2.2 Model Context Uncertainty Analysis

Model context uncertainty is performed using innovative multi method approach

(IMMA) proposed by Mirakyan and De Guio [13] and demonstrated in Singapore
case. The interested reader can find more details in [13].

5.5.2.3 Methods for Model Structure Uncertainty Analysis

Method for the Analysis of Judgmental-Driven Model Structure

Uncertainty in a Targeted Domain
5.5 Multi Method Approaches for Uncertainty Analysis 83

Fig. 5.2  FSUA approach for Use Case-I

84 5 Multi-method Approaches for Uncertainty Analysis

The model structure uncertainty of the judgmental-driven model for Use Case-I is on
conceptual level. The MICMAC judgmental-driven modelling approach is proposed
to implement for Use Case-I for modelling the targeted domain. For the uncer-
tainty analysis of the model structure, the interrelation among different key factors
or descriptors of the study area are presented using Fuzzy linguistic terms such as
weak, moderate or strong influence. Expert elicitation using by group of experts pro-
vides the opportunity to avoid uncertainty in model structure making e.g. judgmental
sensitive analysis. The fuzzy scores also help to incorporate the fuzzy information of
experts, reducing uncertainty because of vagueness at this stage of analysis.
Methods Main function Mixing Mixing Mixing type
rationales sequence
Experts elicitation + Representation Development, Iterative Multi-methodology
fuzzy linguistic of relation extension
terms/fuzzy scores + among ­different
sensitivity analysis key factors in
fuzzy, linguist
terms, scores and
make sensitivity
analysis

Method for the Analysis of Model Structure Uncertainty for Data-Driven

Models in Neighbouring or Distant Domain

Different methods exist for measuring the accuracy of data-driven u­nivariate

models. The accuracy of the prediction depends strongly on the model used.
As discussed in Sect. 5.6.8, Mean Absolute Scaled Error (MASE) and Median
Absolute Percentage Error (MdAPE), is used for data-driven model structure
uncertainty analyses as performance criteria for comparing different models.
Multi model simulation is performed for each individual data set for triangulation,
initiation and evaluation using MASE and MdAPE. Expert elicitation is
implemented for triangulation or initiation as well.

Methods Main function Mixing rationales Mixing sequence Mixing type

EE + MMS + Evaluation Triangulation Iterative, Combination
MASE, MdAPE of model error initiation concurrently
and multi model
simulation

Methods of Model Structure Uncertainty Analysis of Judgmental Driven

Model in Distant Domain

The quality of judgmental-driven models can increase when it is performed

by a group. As stated by Lawrence “[…] the group does produce more accurate
forecasts than simply averaging the individual pre-group judgements” [14]. Expert
5.5 Multi Method Approaches for Uncertainty Analysis 85

elicitation procedure can be implemented in combination with RIM to address

uncertainty related to the categories defined in RIM like trends, weak signals etc.

Methods Main function Mixing Mixing Mixing type

rationales sequence
Expert Group of experts Triangulation Concurrently Combination
elicitations trying to define
different model
entities parallel
and trying to find
consensus

5.5.2.4 Identification of Main Drivers (Key Descriptors)

The MICMAC approach provides not only judgmental modelling of targeted

domain but also to identify key descriptors of the study area. The identification
and ranking of key model descriptors is performed after MICMAC model of study
area is developed. MICMAC is used in combination of expert elicitation.

Methods Main function Mixing rationales Mixing Mixing type

sequence
MICMAC + Identification Extension, Iterative Combination
EE and ranking development
of key descriptors
or factors

5.5.2.5 Methods for Uncertainty Analysis of Models’ Inputs

Methods for the Uncertainty Analysis of Process-Driven Models’ Inputs

in Targeted Domain

The uncertainty of the key model drivers or parameters of process-driven models

for a targeted domain is defined using interval prediction. This is an output of
data-driven model uncertainty analysis. The intervals have been used to define
different future development of key descriptors as a part of scenario analysis.
Fuzzy inference can support the definition of ‘a priory’ probabilities of key
descriptors values and conditional probability among different development of
key descriptors; or else the interdependence among different key descriptors
development can be performed using consistency analysis with linguistic terms.
All method can be used in combination with expert elicitation.
86 5 Multi-method Approaches for Uncertainty Analysis

Methods Main function Mixing rationales Mixing sequence Mixing type

Interval Identification Development Sequential, Combination
predictions of intervals and iterative
quintiles
Fuzzy inference Defining the Expansion, Sequential, Combination
­interrelation development iterative
among different
key descriptors
and ‘a priory’
probabilities
in fuzzy,
linguistic terms
and inference
Expert elicitation Definition possible Expansion, Sequential, Combination
or probable future Development iterative
development of
key descriptors and
their interaction

Methods for Uncertainty Analysis of Data-Driven Models’ Inputs

in Neighbouring Domain

A reference impact matrix (RIM) with expert elicitation and fuzzy inference can
be implemented to address uncertainty analysis of the causal forces, impacted
data-driven models. Inverse modelling approach can be implemented for
identification of model parameter using historical data. Expert elicitation can be
used in combination for development or triangulation.
Methods Main function Mixing rationales Mixing sequence Mixing type
Reference impact Incorporation of Expansion, Sequential, Combination
matrix + fuzzy different impacts development iterative
inference to the model
Inverse modelling Identification Development Iterative Combination
parameter of parameter triangulation
estimation values according
to given data set
of data driven
models
Expert elicitation Identification of Development Sequential, Combination
impact type and triangulation iterative
extent

Methods for Uncertainty Analysis of Judgmental Driven Models’ Inputs

in Distant Domain

RIM use different terms, categories like trends, weak signals, wild cards or
discontinuities as input to model in the distant domain. Uncertainty analysis of
these inputs can be performed using expert elicitation. However it is important to
5.5 Multi Method Approaches for Uncertainty Analysis 87

involve different people with different backgrounds. Furthermore, the involvement

of people without particular expertise in given planning domain can contribute
creative inputs to the modelling and uncertainty analysis.
Methods Main function Mixing rationales Mixing sequence Mixing type
Expert Identification Development Sequential, Combination
elicitation of trends, weak triangulation iterative
signals etc. by different
(non) experts

5.5.2.6 Methods for Uncertainty Analysis of Model Outputs

Methods for Uncertainty Analysis of Process-Driven Models Output

in Targeted Domain

The uncertainty of process-driven models summarises all uncertainty including

model inputs etc. In FSUA, this analysis is performed using scenarios analysis
linked with fuzzy clustering. The model output can be then presented in context of
distinct, consistent and stable scenarios. The scenarios can have fuzzy probability
information if there were enough information or expertise available and fuzzy
inference was implemented to define subjective probabilities before.
Methods Main function Mixing rationales Mixing sequence Mixing type
Scenario Describe how the Expansion Sequential, Combination
analysis system and/or its Iterative
environment could
develop
Representation of
system in consistent,
different, stable
scenarios
Fuzzy Representation of model Development Sequential Combination
clustering output uncertainty
in fuzzy clusters
and identification
of cluster centres

Methods for Uncertainty Analysis of Data-Driven Models Outputs

in Neighbouring Domain

Uncertainty analysis of data-driven model output is presented either quantitatively

by prediction intervals when there are sufficient data available, or qualitatively by
expert elicitation when there is very few or no data available. A combined use of
these approaches is preferable.
88 5 Multi-method Approaches for Uncertainty Analysis

Methods Main function Mixing rationales Mixing sequence Mixing type

Interval Identification Expansion Sequential, Combination
prediction of intervals development iterative
and quintiles
Expert elicitation Identification Development, Sequential, Combination
of interval triangulation iterative
and quintiles

Methods for the Uncertainty Analysis of Judgmental Driven Models

Outputs in Distant Domain

Expert elicitation can be helpful for the uncertainty analysis of models developed
by RIM and using with predefined categories like trends or discontinuities.
Methods Main function Mixing rationales Mixing sequence Mixing type
Reference impact Representation Expansion, Sequential, Combination
matrix of uncertainty development iterative
in different
categories: weak
signal, wild cards
Expert elicitation Assessment Expansion, Concurrently, Combination
of categories, triangulation iterative
identification
of impact type
and extent

5.5.2.7 Model Technical Uncertainty

Model technical uncertainty can be performed using different software for the
same model which requires additional time and resources.
Methods Main function Mixing Mixing Mixing type
rationales sequence
Multi software Computation the Triangulation Concurrently Combination
simulation same model in
­different software

5.5.2.8 Uncertainty Communication

Uncertainty communication is not considered here. However, one important aspect

of uncertainty communication is when presenting the results and discussing with
decision maker. For example, depending on the scale of presented graphs the
perception by decision maker can significantly change, which might have impact to
their decision.
5.5 Multi Method Approaches for Uncertainty Analysis 89

5.5.2.9 Assignment FSUA Methods According Planning and Modelling

Steps for Addressing Different Typologies of Uncertainties

Different methods are used according to different domain and types of uncertainty,
which have been identified and conceptualised in Chap. 4 previously. Some methods
address multiple types of uncertainty and have different roles in different modelling
approaches and domain. For example, interval prediction is used to find the model
output uncertainty of data-driven models; it is also used to determine the model
input uncertainty of process-driven models (Table 5.2).

Table 5.2  FCUA approach, planning and modelling steps

Planning steps -> Planning Phase I Planning Phase II Planning
Phase III
and IV
Modelling steps -> Mental Expresse Conce Formal Com Cal Applied
model d model ptual model puter ibr model
(verbal…) model mode ate
l d
Uncertainty typology mo
del
Context and
IMMA IMMA
framing
Model inputs IP, EE,
and parameters RIM, FI,
IMMA IM
MICMAC,
IM
Epistemic
Model structure EE, SA,
or
FL,
Knowledge
MASE,
uncertainty
MMS,
MdAPE
Model Technical MSS
Model output FC, IP,
EE, SC,
RIM
Vagueness FL x
Linguistic
Ambiguity FL x
uncertainty
Under specificity FL x
Goals/Objectives IMMA IMMA x
Decision- Action/strategies IMMA IMMA x
making Assessment
Uncertainty Criteria
x
Performance
Measure
Communication x x x x x x
Procedural
Available time,
uncertainty x x x x x x x
resources
EE expert elicitation; FC fuzzy clustering; FI fuzzy inference; FL fuzzy linguistic terms; IMMA
innovative multi method approach; IM inverse modelling (parameter estimation); IP interval
prediction; MICMAC matrix of crossed impact multiplications applied to a classification; MASE
mean absolute scaled error; MSS multi software simulation; RIM reference impact matrix; SC
scenario analysis
The colours of the methods have following meaning
Red denotes methods which are for uncertainty analysis of process-driven models in targeted domain
Dark green denotes methods for uncertainty analysis of data-driven models in distant or
neighbouring domain
Blue denotes methods for uncertainty analysis of judgmental-driven models in distant domain
Black denotes methods for the uncertainty analysis of different modelling approaches
90 5 Multi-method Approaches for Uncertainty Analysis

5.5.3 Probabilistic, Random Sampling Based Uncertainty

Analysis (PRSUA) Approach for Use Case-II

5.5.3.1 Analysis Procedure and Steps, Functional View

The specific situation in Use Case-II is that it can have integrated quantitative
model. Assuming that sufficient amount data is available, an approach is
proposed according to the ‘4 analytical sophistication degree’, which is based on
probability theory [1]. Random sampling methods, which are increasingly used
in modelling and planning, are implemented in the PRSUA because of simplicity
and the automatic capability to generate samples, which might reduce subjectivity
issues and efforts working with experts (see Monte Carlo simulation in Sect.
5.6). PRSUA involves several steps and functions. The functions are performed
iteratively as their outputs provide inputs for other functions which in return
may update their inputs. The functions, required resources and constraints with
inputs and outputs for each sub function are presented in the Fig. 5.3. PRSUA
was designed using a similar logic as that for developing the FSUA approach.
The procedure includes five main different functional steps, which require
different inputs and methods to produce results. The main restriction for all of
the functional steps is the availability of data, information, time resources and
expertise. Each box performs certain functions, have input and output information
and require methods and people to perform the function under given constraints.
The methods required for each sub function are discussed in the next sections. A
detailed description of methods is provided in Sect. 5.6.
The functional step model technical uncertainty is not presented in the picture
because its functional position is not fixed. It can be implemented after the 3rd or
4th function.

5.5.3.2 Model Context Uncertainty Analysis

Model context uncertainty can be performed using innovative multi method

approach (IMMA) proposed by [13] and demonstrated in Singapore case. The
interested reader can find more details in [13].
5.5 Multi Method Approaches for Uncertainty Analysis 91

Fig. 5.3  PRSUA for Use Case-II

92 5 Multi-method Approaches for Uncertainty Analysis

5.5.3.3 Methods for Model Structure Uncertainty Analysis

Methods for the Analysis of Process-Driven Model Structure Uncertainty

in a Targeted Domain

Different paradigms exist for the modelling of the processes of real systems, as
discussed in Sect. 3.4.2 in Chap. 3. Modelling in a static equilibrium way for
large energy infrastructure will not provide satisfactory results because of many
interactions, feedbacks and dynamics. As discussed in section “System Dynamic
(SD) Approach to Model the Targeted Zone in Case II—Mexico”, a system dynamics
model can be implemented for modelling a targeted domain. The uncertainty of the
model structure can be divided into two parts: uncertainty in the conceptual model
and uncertainty in the formal model. There are different methods, also called tests
for the uncertainty analysis of system dynamics formal models, such as dimensional
consistency, extreme conditions and behavioural tests. These types of analysis are
particularly important for complex models. Another approach for analysing model
structure uncertainty is multiple model simulation (see Sect. 5.6.9). Modelling the
same real system using different modelling paradigms such as system dynamics and
simple equilibrium approach can prove useful for triangulating results or discovering
inconsistencies. However, using multi model simulation in targeted domain to
model process is very resource intensive. Therefore multi model simulation won’t
be considered in the discussion below. Test for complex model evaluation playing
important role. Depending on the study conditions some test might be implanted
other not. See detailed discussion in Sect. 5.6.12.
The main function, mixing rationales and sequences using these methods in the
overall approach are presented in the table below.

Methods Main function Mixing Mixing Mixing type

rationales sequence
Dimensional Checks each Extension Concurrently, Methodology
consistency model equations iterative enhancement
with respect to
dimensional
correspondence
Extreme conditions Checks the Extension Concurrently, Methodology
and other tests plausible behaviour Triangulation iterative enhancement
of model outputs
while assigning
­different and extreme
values to selected
model inputs
Expert elicitation Judgmental support Development, Concurrently, Combination
and interpretation of triangulation iterative
results of tests
5.5 Multi Method Approaches for Uncertainty Analysis 93

Methods for the Analysis of Model Structure Uncertainty for Data-Driven

Models in Neighbouring or Distant Domain

The methods for the analysis of model structure uncertainty for data-driven m
­ odels
for neighbouring and distant domain modelling are the same as that for FSUA.
Methods Main function Mixing rationales Mixing sequence Mixing type
EE + MMS + Evaluation of model Triangulation Iterative, Combination
MASE, MdAPE error and multi initiation concurrently
model simulation

Methods of Model Structure Uncertainty Analysis of Judgmental Driven

Model in Distant Domain

The quality of judgmental-driven models can increase when it is performed by

a group. As stated by Lawrence “ […] the group does produce more accurate
forecasts than simply averaging the individual pre-group judgements” [14]. Expert
elicitation procedure can be implemented in combination with RIM to address
uncertainty related to the categories defined in RIM like trends, weak signals etc.
Methods Main function Mixing rationales Mixing sequence Mixing type
Expert Group of experts Triangulation Concurrently Combination
elicitations trying to define initiation
different model
entities parallel
and trying to find
consensus

5.5.3.4 Identification Main Model Drivers (Key Descriptors)

The main model drivers or key descriptors are variables that have large impacts on
the model outputs. The sensitive analysis is one of the helpful instruments which
might support this function. However, not only numerical sensitivity (numerical
change of model output value) is important, particularly for complex system
models, but also behaviour or policy sensitivity [15]. As a minimum, to identify
key drivers, the simple on-at-the-time sensitivity analysis method discussed
in Sect. 5.6.11 can be implemented. For representation and ranking of key
descriptors tornado diagram can be very helpful.
Methods Main function Mixing Mixing Mixing type
rationales sequence
Simple, one at Identification and Multiple Sequential Methodology
time sensitivity ranking of key rationales for enhancement
analysis and descriptors of expansion or
tornado diagram targeted domain development
94 5 Multi-method Approaches for Uncertainty Analysis

5.5.3.5 Methods for the Uncertainty Analysis of Models’ Inputs

Methods for the Uncertainty Analysis of Process-Driven Models’ Inputs

in a Targeted Domain

The uncertainty of the key model drivers or variables of process-driven models

for a targeted domain is defined using interval prediction. This is an output of
data-driven model uncertainty analysis. The uncertainty of parameters can be
presented in probability density functions (PDF) if possible. There are different
types of PDFs such as normal, uniform and weibull distributions. If there are
no sufficient data available to define PDFs, then expert judgment can be helpful
fitting a member of a distribution family (normal, exponential, beta, etc.) to those
assessments. It is important to define the right PDFs for model inputs because it
can have a large impact on the model results, as stated by Bukowski et al. [16].
The interdependency of model key descriptors can be defined using correlations
or using copula approach [17] since some correlation among model inputs might
exist. Entropy approach [18] can be also helpful to support experts by definition
of PDFs. Invers modelling can be implemented for calibration as well as expert
elicitation of judgmental support.
Methods Main function Mixing Mixing Mixing type
rationales sequence
Interval Identification of Development Sequential Combination
prediction interval and quintiles
Copula or Defining the Expansion, Sequential, Combination
correlations interdependency development initiative
analysis among key model
descriptors
Inverse ­modelling Identification of Development Iterative Combination
parameter parameter values
estimation according to given
data set
Expert elicitation Definition of PDFs … Development Concurrently, Combination
triangulation iterative

Methods for Uncertainty Analysis of Data-Driven Models’ Inputs

in Neighbouring Domain

The analyses in this step are similar to analysis of FSUA. A reference impact
matrix (RIM) with expert elicitation can be implemented to address uncertainty
analysis of the causal forces influencing data-driven models. Also, inverse
modelling approach can be implemented for identification of model parameters
using historical data. Expert elicitation can be used in combination with data
driven modelling methods for development or triangulation of results.
5.5 Multi Method Approaches for Uncertainty Analysis 95

Methods Main function Mixing Mixing Mixing type

rationales sequence
Inverse modelling Identification of Development Iterative Combination
parameter estimation parameter values
according to given
data set
Reference impact Incorporation of Expansion, Sequential, Combination
matrix and fuzzy different impacts Development iterative
inference to the model
Expert elicitation Identification of Development Sequential, Combination
impact type and triangulation iterative
extent

Methods for Uncertainty Analysis of Judgmental Driven Models’ Inputs

in Distant Domain

Uncertainty analysis of judgmental driven models inputs is preformed similarly

to FSUA see section “Methods for Uncertainty Analysis of Judgmental Driven
Models’ Inputs in Distant Domain”.

5.5.3.6 Methods for Uncertainty Analysis of Model Outputs

Methods for the Uncertainty Analysis of Process-Driven Model Outputs

in Targeted Domain

All of the results of the uncertainty analysis can be summarised by intervals

representing them in the model output via intervals, quantiles etc. The methods
that can be implemented for this task can be sampling techniques like Monte Carlo
Analysis (MCA) (see Sect. 5.6.7).

Methods Main function Mixing rationales Mixing Mixing type

sequence
Monte Carlo Random sampling Expansion Iterative Combination
Analysis (MCA) of uncertain model development
input behaviour
and presenting all
uncertainties in
model output

Methods for the Uncertainty Analysis of Data-Driven Model Outputs

in Neighbouring Domain

The uncertainty analysis of the data-driven model output is similar to FSUA

approach. The interval prediction is developed either quantitatively, when there is
sufficient data available, or, qualitatively by using expert elicitation otherwise.
96 5 Multi-method Approaches for Uncertainty Analysis

Methods Main function Mixing Mixing sequence Mixing type

rationales
Interval prediction Identification Development Sequential, Combination
of interval and iterative
quintiles
Expert elicitation Identification Development Sequential, Combination
of interval and iterative
quintiles

Methods for the Uncertainty Analysis of Judgmental Driven Model Outputs

in the Distant Domain

Expert elicitation can be helpful for the uncertainty analysis of models (trends,
weak signal, discontinuities etc.) developed by RIM and using predefined
categories like trends or discontinuities.
Methods Main function Mixing rationales Mixing sequence Mixing type
Reference impact Representation Expansion, Sequential, Combination
matrix of uncertainty in development iterative
different categories:
weak signal, wild
cards
Expert elicitation Identification Development Concurrently, Combination
of impact type iterative
and extent

5.5.3.7 Model Technical Uncertainty

Expert elicitation can be helpful for the uncertainty analysis of models developed
by RIM and using with predefined categories like trends or discontinuities.
Methods Main function Mixing Mixing Mixing type
rationales sequence
Multi software Computation the Triangulation Concurrently Combination
simulation same model in
different software

5.5.3.8 Uncertainty Communication

Uncertainty communication is not considered here. However, one important aspect

of uncertainty communication is when presenting the results and discussing with
the decision maker(s). Depending on the scale of presented graphs the perception
by decision maker can significantly change, which might impact their decision.
5.5 Multi Method Approaches for Uncertainty Analysis 97

5.5.3.9 Assignment PRSUA Methods According to Planning

and Modelling Steps for Addressing Different Typologies
of Uncertainty

Different PRSUA methods are used according to domain and different types of
uncertainty that were identified and conceptualised in Sect. 4.2, in Chap. 4. Some
methods address multiple types of uncertainty and have different roles in different
modelling approaches and domain. For example, interval prediction is used
for the model output uncertainty of data-driven models and for the model input
uncertainty of process-driven models (Table 5.3).

Table 5.3  PRSUA approach and planning or modelling steps

Planning steps -> Planning Phase I Planning Phase II Planning
Phase III
and IV
Modelling steps -> Mental Express Conce Formal Com Calibra Applied
model ed ptual model puter ted model
model model mod model
(verbal el
Uncertainty typology …)
Context and
IMMA IMMA
framing
IP, SA
Model inputs and
IMMA EE, C, IM
parameters
RIM, FI
MMS,
Epistemic
EE,
or
MASE,
Knowledge Model structure
MdAP
uncertainty
E,
Tests,
Model Technical MSS
MCA,
Model output IP, EE
RIM
Vagueness x x
Linguistic
Ambiguity IMMA x
uncertainty
Under specificity IMMA x
Goals/Objectives IMMA IMMA x
Decision- Actions/strategies IMMA IMMA x
making Assessment
Uncertainty Criteria
x
Performance
Measure
Communication x x x x x x
Procedural
Available time,
uncertainty x x x x x x x
resources
C correlation or copula; EE expert elicitation; FI fuzzy inference; IMMA innovative multi method
approach; IM inverse modelling (parameter estimation); IP interval prediction; MASE mean
absolute scaled error; MdAPE median absolute percentage error; MCA Monte Carlo analysis;
MMS multi model simulation; MSS multi software simulation; RIM reference impact matrix; SA
sensitivity analysis; Tests different tests for complex system models such as extreme conditions
or dimensional consistency test
The colours of the methods have following meaning
Red denotes methods which are for uncertainty analysis of process-driven models in targeted domain
Dark green denotes methods for uncertainty analysis of data-driven models in distant or
neighbouring domain
Blue denotes methods for uncertainty analysis of judgmental-driven models in distant domain
Black denotes methods for the uncertainty analysis of different modelling approaches
98 5 Multi-method Approaches for Uncertainty Analysis

5.6 A Review of Methods and Methodologies

for Uncertainty Analysis

5.6.1 Correlations and Copulas

5.6.1.1 Description

Clemen and Reilly [19] describe three correlation assessment methods, which can
support experts who think about the relation of random variables. In a statistical
approach, an expert might make a judgment based, e.g., on R2 that would result
from the regression of one variable against another. An alternative possibility
is for experts to work with several scatterplots representing different levels of
correlation and select one that is consistent with their belief regarding the strength
of the relationship between the variables. Another approach is concordance
probability. Here, an assessment of conditional or joint probabilities is related to
Pc concordance probability. For a bivariate population (X, Y), by considering two
independent draws (x1, y1) and (x2, y2) Pc can be defined [19] e.g.
Pc = P(x1 ≤ x2 |y1 ≤ y2 )
The third approach is conditional Fractile Estimates. Here Spearman’s ρ is derived
using conditional estimates. Let F(x) and G(y) be distribution functions for X
and Y random variables respectively, then the standard nonparametric regression
representation is [19]:
 
E[F(X)|y] = ρXY G(y) − 0.5 + 0.5
where ρXY is the Spearman correlation between X and Y.
The common way of defining conditional distributions among different model
input variables is a demanding task because the amount of input variables in
large city models, such as infrastructure development, are very high. The experts
or analysts have to define each variables interaction probability distributions, the
number of which can be very high. An alternative approach for construction of
joint distributions is using copula. Copula-based joint distribution can be defined
using rank-order correlations to reduce the number of required assessments and
relax the need to search for conditional independence [19]. In the copula approach,
a joint distribution of random variables is expressed as a function of the marginal
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 99

distributions. Sklar’s theorem is the building block of the theory of copulas [17]:
Let xn be a random variable with F1 (x1 ), F2 (x2 ) . . . Fn (xn ) marginal distribution
functions. Then there exist a copula function C such as for all (x1 , x2 , . . . xn ) joint
cumulative distribution function F(x1 , x2 , . . . xn )
F(x1 , x2 , . . . xn ) = C(F1 (x1 ), F2 (x2 ) . . . Fn (xn ))
C is uniquely determined when F1 (x1 ), F2 (x2 ) . . . Fn (xn ) are all continuous.
There are different ways to construct and use copula-based joint densities
such as multivariate normal copulas or maximum-entropy copulas discussed by
Abbas [20].

5.6.1.2 Typology of Uncertainty Addressed

C is appropriate for model inputs uncertainty in risk conditions for process driven
models.

5.6.1.3 Potential, Main Rationales

These methods are implemented for extension and addressing interdependency

among random variables. They can be used also for development to inform other
methods such as the Monte Carlo simulation.

5.6.1.4 Performance According to Some Quality Factors

Technical Quality Factors

• Extension of incorporation of uncertainty typologies: See discussion in

Sect. 5.6.1.2.
• Incorporating qualitative and quantitative information. ‘C’ approaches use only
quantitative information.
• Automatism. Correlations and copula can be automatized using formalised
procedures.
• Mixing synergy: see the main rationales above.

Organisational Capability

• Required time and resources: Main time and resource requirements are linked
with the work of experts, which can be defined as mid- to high-time or resource
requirements.
• Required information: Information needs can be on the level “Int05”.
• Required expertise and skills: If using C, familiarity with statistical and
probability concepts are required.
100 5 Multi-method Approaches for Uncertainty Analysis

• Flexibility: C is appropriate, particularly in the planning phase II in quantitative

modelling. It can be used in the Use Case-II for the targeted domain.
• Legitimacy: No information is available at this stage of research.

Satisfaction by Planning Participants

• Satisfaction working with the method: Familiarity with statistical and probability
concepts related to correlations etc. is not always the case by planners and
analysers having an engineering background.
• Satisfaction by DM with the method: No information is available at this stage of
research.

5.6.1.5 Future Reading

Theoretical discussion, empirical applications and recent achievements in the field

of Copula are discussed in [17]. An overall introduction about copulas is presented
in [21].

5.6.2 Expert Elicitation

5.6.2.1 Description

Formal expert elicitation (EE) is a structured approach to systematically consult

experts on uncertain issues1 [22]. EE is widely used technique in energy planning
and modelling. However, there was not discussion found about systematically
formalised procedure about this method used in cities and territory energy
planning (see review in Chap. 3). In general, EE is used to quantify ranges for
poorly known parameters representing sometimes with subjective probability
density function or for estimation of most expected value for model design.
Different elicitation protocols have been developed and implemented like
Stanford/SRI Protocol [23]. Knol et al. [22] present seven step towards organising
EE in the context of environmental health impact assessment. They are: (1)
Characterisation of uncertainties (2) Scope and format of the elicitation (3)
Selection of experts (4) Design of the elicitation protocol (5) Preparation of the
elicitation session (6) Elicitation of expert judgements (7) Possible aggregation
and reporting. These steps are generic and can be implemented also in energy
planning and modelling context.

1Experts in this context are persons having specific knowledge about certain domain e.g. in

demography or energy technology.

5.6 A Review of Methods and Methodologies for Uncertainty Analysis 101

5.6.2.2 Typology of Uncertainty Addressed

Expert judgment using EE is and can be implemented particularly for scenario or

ignorance located in any modelling process and uncertainty type.

5.6.2.3 Potential, Main Rationales

The method is usually implemented in the situation where the availability of data or
information is low in order to develop information for planning or for other methods.
It is also used concurrently for triangulation or validation of empirical methods.

5.6.2.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.2.2.

• Incorporating qualitative and quantitative information. EE can use both
information.
• Automatism. EE has usually low automation. However, the procedure can be
formalised and standardised using elicitation protocols or interactive computer
software.
• Mixing synergy : see main rationales above.

Organisational Capability

• Required time and recourses: EE procedure is high time and recourse intensive.
E.g. expert identification, involvement and elicitation procedure.
• Required information: EE methods do not require high data intensity, it can be
on Level “Int 01” but also higher depending with which method is EE used.
• Required expertise and skills: No particular expertise or skills are required to
use EE.
• Flexibility: EE is appropriate in any planning and modelling context and
planning phase.
• Legitimacy: Legitimacy of EE depends on special city or territory conditions
and availability of domain experts.

Satisfaction by Planning Participants

• Satisfaction working with method: In most of cases EE is performed without

special organised procedure. Future research are needed to evaluate satisfaction
using e.g. elicitation protocols in EE.
102 5 Multi-method Approaches for Uncertainty Analysis

• Satisfaction by DM with method: For decision maker (DM) the certain formalisation
of method which helps to incorporate domain knowledge of experts can increase
satisfaction of DM.

5.6.2.5 Future Reading

Large list of papers, books, example studies and software are presented by [11, 24–26].

5.6.3 Fuzzy Inference

5.6.3.1 Description

Fuzzy inference is based on compositional rules of inference. There are two different
fuzzy inference techniques: Mamdani and Sugeno-style. Mamdani method is more
intuitive and user friendly in contrast Sugeno-style is more computationally efficient
[27]. The process of Mamdani knowledge based fuzz-inference has different steps:

Fuzzification
First step is to determine which crisp inputs belong to which appropriate fuzzy
stets. The crisp inputs can be a priory probability of values of key factors or
conditional probability used e.g. in a cross impact (CIM) analysis among different
values of different key factors. Here the formulation of e.g. a priory probability is
not any more to say the probability of occupancy of certain value of key factor is
P = 0.15 but it is defined now in fuzzy linguistic terms that the occupancy is “very
seldom”. The idea is to replace crisp probabilities through linguistic variables
and membership functions. The similar procedure is performed for conditional
probabilities using in CIM.
For definition membership values of crisp sets trapezoidal membership function
is commonly used.

Fuzzy inference: knowledge-based Fuzzy rules system

The expert knowledge about the relationships of the input and output ​​values is
modelled with a set of if-then rules, which is the real expertise in knowledge-based
Fuzzy rules system. All if-then rules to be together are named as a knowledge-base.
Rules consist of two parts: IF part, called the antecedent and the THEN part
called the consequent.
The basic syntax of a rule is:
IF<antecedent>
THEN<consequent>
In general, a rule can have multiple antecedents joined by the keywords AND
(conjunction), OR (disjunction) or a combination of both. For Intersection AND-
operator [28].
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 103

µA∩B (x) = min[µA (x), µB (x)] = µA (x) ∩ µB (x)] where x ∈ X

Every combination is a rule from two premises and one conclusion. The rules
must be combined aggregating all outputs of rules. Aggregation is the process
where fuzzy sets which represent the outputs of each rule are combining into
a single fuzzy set. The rules can be also weighted before combining. There are
different implication methods e.g. Max (maximum AND), prober (probabilistic
OR) or sum (simply sum of each rules output set).

Defuzzification
After fuzzing input values and using rules evaluation forming linguistic
variables and aggregating them, now membership values have to be again
translated into crisp values. The aggregated fuzzy set has range of output values
and must be defuzzified. There are several methods for evaluation of a single
defuzzified number. Most commonly used methods are centroid method Centre
of Gravity (CoG), the average maximum method or weighted average maxima
method [29].

5.6.3.2 Potential, Main Rationales

The method is usually implemented in the situation where the availability of data
or information is low or no precise to develop information for planning or for other
methods. It can be also used concurrently for triangulation with other methods.

5.6.3.3 Typology of Uncertainty Addressed

FI is appropriate to address linguistic uncertainty particularly uncertainty because

of ambiguity.

5.6.3.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.3.3

• Incorporating qualitative and quantitative information. Fuzzy inference can use
both information, quantitate, linguistic or qualitative information.
• Automatism. Fuzzy inference has usually middle level of automation. The
procedure is well formalised and scandalised.
• Mixing synergy: See main rationales above.
104 5 Multi-method Approaches for Uncertainty Analysis

Organisational Capability

• Required time and recourses: FI procedure is high time and recourse intensive
only for identification, involvement and elicitation of experts. If the input values
are available FI has low time or recourse intensity.
• Required information: EPR methods do not require high data intensity, it can be
on Level “Int 01” to “Int 03”.
• Required expertise and skills: Certain expertise or skills are required for using
FI, which is not included in common educational systems.
• Flexibility: FI is appropriate for all planning phase. Use Cases and modelling
domains.
• Legitimacy: Legitimacy of FI depends on special city or territory conditions.

Satisfaction by Planning Participants

Satisfaction working with method: FI supports better facilitation of expert

knowledge using qualitative and quantitative information which increase
satisfaction working with method.
Satisfaction by DM with method: Increasing incorporation of different qualitative
and quantitative information with low costs and nevertheless presenting the
information in numerical values might increase satisfaction level of DM. However,
future research is needed to evaluate satisfaction using FI.

5.6.3.5 Future Reading

Information about the method can be founded in [30–32].

5.6.4 Innovative Multimethod Approach (IMMA)

5.6.4.1 Description

IMMA methodology combines four different methods from soft operational

research and from the area of inventive problem solving. The methodology is
designed to address different task in planning preparation and orientation phase
like for addressing model context uncertainty, for definition of conceptual model,
objectives, solutions or barriers. Additionally IMMA makes possible to identify
the contradictions in given planning context. It suggests also ways how to
overcame or solve these contradictions.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 105

5.6.4.2 Typology of Uncertainty Addressed

IMMA can address different types of uncertainty. For dealing with uncertainties
related to context framing and conceptual model development or to objective and
solutions definition IMMA can provide valuable help.

5.6.4.3 Potential, Main Rationales

IMMA can be used for multiple rationales. It can be helpful e.g. for development to
generate inputs for other methods like conceptual model or for extension addressing
issues which cannot be addressed using other methods in given planning context.

5.6.4.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.4.2.

• Incorporating qualitative and quantitative information. IMMA can use both
information quantitative and qualitative.
• Automatism. IMMA has low automation. However, the procedure is well formalised
and standardised using methodological steps, standard questions and graphical tools.
• Mixing synergy : See main rationales above.

Organisational Capability

• Required time and recourses: IMMA procedure is high time and recourse
intensive. E.g. key stakeholder involvement or organisation of workshops.
• Required information: IMMA methods do not require high data intensity, it can
be on Level “Int 01”.
• Required expertise and skills: No particular expertise or skills are required for
IMMA.
• Flexibility: IMMA is appropriate particularly at the beginning of planning and
modelling. It can be implemented in both Use Cases.
• Legitimacy: Legitimacy of IMMA depends on special city or territory conditions.

Satisfaction by Planning Participants

• Satisfaction working with method: IMMA have not been used frequently in
different studies.
• Satisfaction by DM with method: Standardised questions, procedure with steps
and graphical interactive maps might provide satisfaction by decision maker.
However, future studies are required for performance assessment of IMMA.
106 5 Multi-method Approaches for Uncertainty Analysis

5.6.4.5 Future Reading

IMMA is presented, and discussed in [13]. In the same study the applicability of
the methodology is demonstrated and the performance of IMMA is evaluated.

5.6.5 Inverse Modelling
5.6.5.1 Description

Inverse modelling (IM) or also denoted as automatic calibration are used for
parameter estimation. Parameter values are calibrated through optimisation. Objective
function is minimizing in order to find set of optimal parameters values. Most of
these techniques have the benefit that they produce calibration statistics in terms of
parameter correlation and parameter uncertainties as well. However, the particular
limitation of this method is that the model calibration is based on a single model.
Parameter uncertainty does not include in this case uncertainty due to model structure.
[10]. IM can be implemented only for process- and data-driven models [33].

5.6.5.2 Potential, Main Rationales

The method is usually implemented in the situation where the availability of

data or information is low to generate information, development for planning or
for other methods. It is also used concurrently for triangulation or validation of
empirical methods.

5.6.5.3 Typology of Uncertainty Addressed

IM address model input parameter uncertainty.

5.6.5.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.5.3.

• Incorporating qualitative and quantitative information. Only quantitative
information can incorporated IM.
• Automatism. IM has high automation.
• Mixing synergy: see main rationales above.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 107

Organisational Capability

• Required time and recourses: IM require very low time and recourses if data for
calibration of model parameters are available.
• Required information: IM methods require high data intensity, it can be on
Level “Int 04”.
• Required expertise and skills: Some expertise or skills about optimisation
approaches might be required.
• Flexibility: IM is appropriate particularly in planning phase II in quantitative
modelling and planning phase, in can be used only in Use Case-II for targeted
domain.
• Legitimacy: Legitimacy of IM depends on special city or territory conditions.

Satisfaction by Planning Participants

• Satisfaction working with method: No information is available at this stage of

research. However, in some own studies we were satisfy with method.
• Satisfaction by DM with method: No information is available at this stage of
research.

5.6.5.5 Future Reading

Interesting references about IM are mentioned by [10].

5.6.6 Interval Prediction (IP) in Data Driven Models

5.6.6.1 Description

Chatfield [34] defines prediction interval or interval prediction (IP) as “IP is an

estimate of an (unknown) future value that can be regarded as a random variable
at the time the forecast is made” Interval forecast consist of an upper and lower
limit between which future value expected to be: It can be presented by different
probabilities or intervals e.g. 5, 25, 75 or 95 %.
Different approaches exist to compute interval prediction. it can also be used ad
hoc IP method based on the empirical properties of the residuals [35].
Most common PI calculation method in univariate modelling is theoretical
formulae conditional on a best-fitting model which depends on forecasting model
and normality assumption. Other methods which are less depended on forecasting
fitted model are empirically based [36] or resampling methods. IP calculation
general method is

x̂n ± q α Var(et (h)),
2
108 5 Multi-method Approaches for Uncertainty Analysis

(1-α)-Prediction-intervals is built with the forecast mean value x̂T+h added with
the α2- quintile of the normal distribution multiplied with the root of the variance of
the forecast error et (h) [34].
In some situation when not enough observation is available and there are
doubts about model assumptions empirical based or resampling methods can be
considered [34].
Next implemented method for calculating IP is resampling method: bootstrapping
which do not need a priory assumption about parametric distribution (usually
normal). It samples from the empirical distribution of past fitted errors [34]. For
complex models or for small data set bootstrapping is particularly helpful [34]. The
basic idea is: if we cannot sample from the entirely dataset (e.g. not enough data or
distribution is not known), the sampling is done from approximation of total data
set. Some observations from original set might occur more than once in bootstrap
samples and some not at all.
Another method for calculating IP is judgmental IP As stated by Wright et al.
[37] on the base of review of different findings “Evidence on the accuracy and
calibration of judgmental interval prediction is not very encouraging.” Reviewing
several study [37] found that judgmental IP is influenced by data characteristics
like seasonality, trend, or randomness. It is also stated that even scale of a
graphical plot presented for judgment can influence to define IP [37].
In general [34] states that empirical studies have shown that judgmental PI’s
tend to be too narrow on average. The reason for that can be very different e.g.
there are outliners in the data, the wrong model may be used or the model may
change either during the period of fit or in the future [34]. Because of that [38]
suggests a modification of equation above √
Var(et (h))
x̂T+h ± √
∝
√
Where the constant 1/ ∝ (which replaces q α) is selected using an argument based
2
on Chebychev’s inequality.
Specific interval generation approaches for time series analysis methods like
exponential smoothing was proposed by Hyndman et al. [39] and for ARIMA
approaches by Snyder et al. [40].

5.6.6.2 Potential, Main Rationales

The method is usually implemented in the situation where the availability of

data or information is low to generate information, development for planning or
for other methods. It is also used concurrently for triangulation or validation of
empirical methods.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 109

5.6.6.3 Typology of Uncertainty Addressed

IP addresses mainly parameter uncertainty. However, it can be used for other

uncertainty like representing model structure errors using intervals etc.

5.6.6.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in

Sect. 5.6.6.3.
• Incorporating qualitative and quantitative information. For IP developed by
experts qualitative and quantitative information can be used. Other methods for
IP development require quantitative information.
• Automatism. IP has high automation, IP generated by experts not.
• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: Only qualitative IP require high time and
recourses for involvement of experts. Other quantitative IP methods have low
time and recourses needs.
• Required information: Qualitative methods for IP do not have high data
intensity. Quantitative IP generation methods require high data intensity, it can
be at least on Level “Int 02”.
• Required expertise and skills: Some expertise or skills about statistic
approaches might be required
• Flexibility: IP is appropriate particularly in planning phase II in quantitative
modelling and planning. It is appropriated for neighbouring domain for both
Use Cases.
• Legitimacy: Legitimacy of IP depends on special city or territory conditions. No
information is available at this stage of research.

Satisfaction by planning participants

• Satisfaction working with method: Implementing IP gives satisfactory results in

both Use Cases.
• Satisfaction by DM with method: Generated intervals using IP methods is more
accepted by DM than describing uncertainty in other statistical terms.
110 5 Multi-method Approaches for Uncertainty Analysis

5.6.6.5 Future Reading

Detailed discussions about different IP methods are presented in [34, 35, 39].

5.6.7 Monte Carlo Simulation

5.6.7.1 Description

Monte Carlo Analysis (MCA) is a statistical sampling technique in obtaining a

probabilistic approximation to the solution of a mathematical equation or model.
Its purpose is to trace out the structure of the distributions of model output
those results from specified uncertainty distributions of model inputs and model
parameters [10]. MCA can be also used to identify key sources of uncertainty
and to quantify the relative contribution of these sources to the overall variance
and range of model results. To reduce the required number of model runs and
computation time, new sampling methods have been designed such as Latin Hyper
Cube sampling. 16 different application steps are suggested by Van der Sluijs et al.
[11] as a guideline like (1) Conduct preliminary sensitivity analyses, (2) Restrict
the use of probabilistic assessment to significant parameters or (3) Use data to
inform the choice of input distributions for model parameters.

5.6.7.2 Potential, Main Rationales

The method is usually implemented in the situation where the availability of data
or information is low to develop information for planning or for other methods. It
can be also used concurrently for triangulation.

5.6.7.3 Typology of Uncertainty Addressed

MCA usually address statistical uncertainty in model inputs drivers and

parameters. It could be used also for model structure uncertainty. Two steps MCA
can help to separate treatment of epistemic and aleatory uncertainty [11].

5.6.7.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.7.3.

• Incorporating qualitative and quantitative information. MCA use only
quantitative information.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 111

• Automatism. MCA has high automation, generating random samples.

• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: MCA method has low time and recourses
requirements.
• Required information: MCA methods have high data intensity, it can be up to
Level “Int 05”.
• Required expertise and skills: Some expertise or skills about statistic approaches
might be required.
• Flexibility: MCA is appropriate particularly in planning phase II in quantitative
modelling. It can be only used in Use Case-II for targeted domain.
• Legitimacy: Legitimacy of MCA depends on special city or territory conditions.
No information is available at this stage of research.

Satisfaction by Planning Participants

• Satisfaction working with method: Automatic quality of MCA makes possible

to perform different experiments giving satisfactory results. However, sufficient
data must be available.
• Satisfaction by DM with method: No information is available at this stage of
research.

5.6.7.5 Future Reading

Large amount different books and studies exist about MCA. References [10, 11]
give short description of MCA. For more detailed information [41] can be useful.

5.6.8 Multiple Model Simulation (MMS) of Process

Driven Models

5.6.8.1 Description

Model structure uncertainty depends not only on functional form of dynamic

interrelation among several parameters and variables of the model or through dis-/
aggregation of different system components in time and space but also the way of
presenting complexity using different modelling paradigms like system dynamics
or behavioural driven models.
Instead using single model Multiple Model Simulation (MMS) can be
implemented using different models e.g. using different conceptual models with
112 5 Multi-method Approaches for Uncertainty Analysis

process description or different modelling paradigms like system dynamics

vs. static equilibrium models or comparing process driven models like system
dynamic model with data driven models like neural nets.
Complex system modelling required additional performance assessment or tests
which will be discussed in the Sect. 5.6.12.

5.6.8.2 Potential, Main Rationales

The method is usually implemented in the situation where the availability of

data or information is low to generate information, development for planning or
for other methods. It is also used concurrently for triangulation or validation of
empirical methods.

5.6.8.3 Typology of Uncertainty Addressed

MMS usually address model structure uncertainty.

5.6.8.4 Performance According to Some Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.7.3.

• Incorporating qualitative and quantitative information. MMS can use quantitative
information.
• Automatism. It depends on methods used.
• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: MMS method has very high time and recourses
requirements for process driven modelling because for the same real system
two different methods must be implemented for targeted domain. MMS for data
driven modelling has low time and recourses requirements.
• Required information: Data intensity depends on MMS methods used.
• Required expertise and skills: Expertise or skills about different modelling
approaches like complex system modelling and static equilibrium modelling are
required. Also using different data driven modelling approach requires expertise
and skills.
• Flexibility: MMS is appropriate particularly in planning phase II in quantitative
modelling. In can be used in Use Case-II in for targeted domain.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 113

• Legitimacy: Legitimacy of MMS depends on special city or territory conditions.

No information is available at this stage of research.

Satisfaction by Planning Participants

• Satisfaction working with method: The planner and analysis were satisfied using
MMS system dynamics and stationary equilibrium models. However, in other
situation it can be different because of high time and resource requirements.
• Satisfaction by DM with method: No information is available at this stage of
research.

5.6.8.5 Future Reading

A new framework for coping with model structure uncertainty bases on different
conceptual model and assessment of their pedigree is presented by Refsgaard et al.
[42]. Using parallel modelling [43] could identify key uncertainty in one of sub models.

5.6.9 Multiple Model Simulation (MMS)

of Data Driven Models

5.6.9.1 Description

Different methods exist for MMS for data driven modelling. Different criteria
are used for MMS comparison of data driven models, particularly for time
series models for prediction. Making a review [44] propose “we prefer Mean
Absolute Scaled Error (MASE) as it is less sensitive to outliers and more easily
interpreted than Root Mean Squared Scaled Error (RMSSE), and less variable
on small samples than Median Absolute Scaled Error (MdASE)”. Alternatively
Median Absolute Percentage Error (MdAPE) and Geometric Mean of the Relative
Absolute Error (GMRAE) are also recommended as a criteria by Armstrong and
Collopy [45]. They have high reliability for annual based data forecasting for
longer period.
MdAPE and MASE will be considered in the analysis.
The absolute percentage error (APE) for given horizon of particular series is
defined as
 
 et 
APEt =  
Yt

and MdAPE
MdAPE = medianAPEt
114 5 Multi-method Approaches for Uncertainty Analysis

Average scaled error (ASE)

et
ASEt = 1 n
(n−1) i=2 |Yi − Yi−1 |

et is forecast error,Yi denote the observation at time tYi−1 denote the observation at
time t-1n are time periods consideredand MASE
MASE = mean|ASEt |

5.6.9.2 Potential, Main Rationales

The method is usually implemented in the situation where different models have to be
compared. It is used concurrently for triangulation or validation of empirical methods.

5.6.9.3 Typology of Uncertainty Addressed

MMS usually address model structure uncertainty.

5.6.9.4 Performance According to Some Quality Factors

Technical quality factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.9.3

• Incorporating qualitative and quantitative information. MMS can use only
quantitative information.
• Automatism. It is highly automatized once the scripts are developed.
• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: MMS method do not have high time and
recourses requirements because for the same data stets different methods can
be implemented. Particularly, if there are sufficient data amount available using
some automatic methods like auto ARIMA for secondary method for MMS the
time and recourses requirement is very low.
• Required information: Data intensity depends on data driven modelling
approach and not on MMS.
• Required expertise and skills: Some expertise or skills about different modelling
approaches like neuronal nets or statistical approaches are required as well as
for different MMS criteria.
• Flexibility: MMS is appropriate particularly in planning phase II in quantitative
modelling, for neighbouring domain for both Use Cases.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 115

• Legitimacy: Legitimacy of MMS depends on special city or territory conditions.

No information is available at this stage of research.

Satisfaction by planning participants

• Satisfaction working with method: The planner and analysis were satisfied
using MMS if sufficient data is available for using data driven modelling.
• Satisfaction by DM with method: No information is available at this stage of
research.

5.6.9.5 Future Reading

A survey about of evaluation criteria for data driven extrapolated models and their
relative importance as determined academicians and practitioners is done by [46].
A discussion about comparison of measures of accuracy is provided in [44].

5.6.10 Scenario Analysis and Fuzzy Clustering

5.6.10.1 Description

A scenario is a consistent, different, stable, plausible description of how

the system and/or its environment may/could/should develop in the future.
Scenarios have different functions in planning and modelling e.g. explorative
or explanatory function. It can serve as a background for decision making and
strategy formulation or for supporting communication and learning. Large
variety of scenario categories exists e.g. normative scenarios, descriptive
scenarios, explorative or anticipative scenarios. Analysing and comparing
different categories of scenarios in context of energy planning Zürni [47]
indicates several advantages using explorative, descriptive scenarios. In contrast
to normative scenarios descriptive scenarios are developed on the base of cause
and effect relation and there are no normative judgments of overall scenarios by
the user and creator. Descriptive scenarios are also explorative where a scenario
building starts in the present and explores possible projections into the future.
Main scenario development steps are:
• Scenario field identification
• Key descriptors or factors identification
• Key descriptors analysis
• Scenario generation: raw scenarios
• Scenario selection and definition
116 5 Multi-method Approaches for Uncertainty Analysis

Step 1

Different sub methods can be implemented in each step. For step 1, for field
identification or problem framing different methods can be implemented which are
discussing in Use Case-I.

Step 2

For step 2, for key descriptors or factors identification one at time sensitivity
analysis can be used for Use Case-II or MICMAC (Matrice d’Impacts Croisés -
Multiplication Appliquée à un) or system grid can be implemented for Use Case-I.

Step 3

Key descriptor analysis can be performed either using qualitative judgmental

methods or data driven time series methods with interval prediction.

Step 4

In these step interaction among different factors must be defined either using
consistency analysis or cross impact analysis. The final set of factors values will
be evaluated accordingly with total consistency or probability measure.
Consistency numbers are subjective, ordinal, contradiction free relationship
between two variables/key factors values. They describe interdependency.
The interdependency can be presented in way to show whether to variables are
supportive intensifying the direction of development, opposing or neutral [48].
An example building consistency matrix is demonstrated below. Consistency
analysis is carried out for each pair of key model input values/factors [48]. All
factors have to be checked with each other, values Ai) or Bj) of key factors “A”
and “B” are compared. The consistency of each comparison is assessed using
different scales e.g. (1–5) or (−2, −1, 0, 1, 2) [48], −2 = strong inconsistency,
−1 = weak inconsistency, 0 = neutrality, 1 = weak consistency, 2 = strong
consistency (Table 5.4).

Table 5.4  An example of consistency matrix

5.6 A Review of Methods and Methodologies for Uncertainty Analysis 117

This is not complete enumeration, so only the half of table needed to be filed
in. The information for probability is not needed in consistency analysis.
Consistency values can be summarised in additive approach as follow:

 N−1
 N

nj nj
Csum (Sm ) = c(kini · kj ) = c(kini · kj )
Sm:ni,nj i=1 j=i+1

K Key factor or descriptor

nj
kj A value of key factor kj
nj nj
c(kini · kj ) Consistency value of combination value kj of Kj key factor and value
ki of Ki factors
ni

Csum (Sm ) Sum of consistency values of different combinations for Smscenario

(CIM) Cross-Impact analysis is a group of analysis approaches which accounts the
conditional probable relationships among of different possible values of key factors.
It is used to analyse plausibility of scenarios. “Plausibility” here means probability.
In contrast to consistency analysis, cross-impact analysis is full enumerated,
all interdependency among values of factors have to be computed. CIM analysis
can be divided into tree group: correlated, causal-static and causal-dynamic [49].
Below causal-dynamic approach is discussed using cross-impact computation
method familiar to Battelle-Institute in Frankfurt am Main approach implemented
in SzenoPlan software (Table 5.5).
The probability of each comparison is assessed using different scales e.g. (1–7) or
(−3, −2, −1, 0, 1, 2, 3), −3 = extreme reduction, −2 = strong reduction, −1 = weak
reduction, 0 = neutral, 1 = weak growing, 2 = strong growing, 3 extreme growing.
Initial probability (a priory) of an “event” (key factor value) is estimated for
each factor. This is done by each event independency from each other.
In next step conditional probabilities is defined asking the question e.g. “If
event Bi) occurs, how will it change probability of occupancy of Ai)?” In the
case see above table it is −3, which means probability occurrence of Ai) will be
reduced by coefficient computed using −3 value from table. Before using these
coefficients for defining posterior probabilities, following classification for these
coefficients are using (Table 5.6).

Table 5.5  An example of cross-impact matrix

118 5 Multi-method Approaches for Uncertainty Analysis

Table 5.6  CIM and Cross-impact value −3 −2 −1 0 1 2 3

coefficients
Coefficient 1/4 1/3 1/2 1 2 3 4

New probability is evaluated as follow

Coefficent ∗ Pprior
Pnew =
1 + Pprior (Coefficent − 1)
Pprior is a priory probability of an event of key factor value defined before without
cross impactPnew is new probability of the event after each iteration with cross impact
1
Pnorm = n ∗ Pnew
i=1 P(i)
Pnorm is the normed new posterior probability of an event or key descriptor value
taking into account cross impact.
Another approach for defining interaction among different factors and building
scenarios can be Fuzzy-inference discussed in Sect. 5.6.3 which incorporated also
linguistic uncertainty.

Step 5 Scenario Selection

Different methods exist to identify and select final scenarios like according to high
consistency or probability value. However, scenarios must be different, stable,
consistent and plausible set of factors or descriptors [50]. One of the methods
supporting this function is clustering approach. The purpose of the cluster analysis is to
divide objects (in our case values of descriptors) in a groups or clusters in a way that:
the objects of a group are as much as possible similar homogeneous to each other and
the objects from different clusters are as much as possible different, heterogeneous.
Assignment of certain partition to exactly one cluster using classical clustering
method often leads to loss internal homogeneity within a cluster and loos of
information. The fuzzy cluster analysis has on the other hand the opportunity to deal
with this problem using membership function and assigning membership to each object
to each cluster. This will help also to identify objects with high membership value in
cluster and which might be the potential representative scenario for given cluster.
The Fuzzy C-Means algorithm (FCM) [51] is the most common used fuzzy
clustering algorithm. It is objective Function Method which allow the most precise
(but not necessarily more valid) formulation of the clustering criterion [51]. The
task to find fuzzy clusters in a space can be done minimizing target function of the
problem. The function J (X, U, B) is formulated as:
n 
 c  
J(X, U, B) = m
ui,j · d 2 β�i , x�j
j=1 i=1
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 119

With
c Amount of clusters
n Amount of objects,
m fuzzy exponent m ∈ (1,∞),
ui,j degree of membership of object j to cluster i,
d Distance between object and cluster centre i,
xj Character vector of objects j, X = {�x1 , x�2 , . . . , x�n }
 
βi Character vector of cluster centre i, B = β�1 , β�2 , . . . , β�c
The function must be minimize taking into account following restrictions
c

ui,j = 1 for all j ∈ (1, . . . , n)
i=1
n

ui,j > 1 for all i ∈ (1, . . . , c)
j=1

A direct solution of the optimization problem is not possible therefore the

objective function is minimized iteratively [51]. Fuzzy clustering is carried out
through an iterative optimization of the objective function shown above, with the
update of membership ui,j, and the cluster centres zj.
 1

 � 2 � 1 , falls Ij = ∅,
 �
 c d ( β�i ,�xj ) m−1
k=1 d 2 β� ,�x
ui,j = ( k j)

 0 falls Ij = ∅ and i ∈
/ Ij ,

 �
x, x ∈ [0, 1], as i∈Ij ui,j = 1gilt, falls Ij = ∅ and i ∈ Ij ,
 
in which Ij = (1 ≤ i ≤ C, d2 β� i , x�j = 0)
n m�
j=1 ui,j x j
�zi = n m
u
j=1 i,j

The starting amount of cluster is not known therefore it can be selected e.g. from 2
to 9 cluster. The final amount of scenarios depends on cluster qualities or validity.
There are different quality or validity criteria’s like the Partition Coefficient,
Classification Entropy, Proportion Exponent, Normalization and Standardization
of Validity Functional etc. [51].
The simplest global quality criterion is to be minimized objective function J
(X, U, B). However, in order to reduce the large number of cluster and provide
some quality measure criteria like portion coefficient or classification entropy can
be implemented.
Partition coefficient or degree of separation
The idea based on that a good clustering of the objects makes possible the objects
clearly assign to certain cluster. The membership value must be close to 1 or
120 5 Multi-method Approaches for Uncertainty Analysis

respectively to 0. This criterion is aimed to assign objects to clusters clearly; a

higher value indicates a better classification.
c n
1  2
PC(U) = ui,j
n
i=1 j=1

Classification Entropy
Classification Entropy is an indicator for internal homogeneity of cluster. The
quality of clustering is bigger if entropy is smaller.
c n
1 
PE(U) = − ui,j ln(ui,j )
n
i=1 j=1

5.6.10.2 Potential, Main Rationales

The method is usually implemented in the situation to develop information for

planning or for other methods. It is also used for extension.

5.6.10.3 Typology of Uncertainty Addressed

SC typically addressee ignorance and uncertainty of model input parameters

and driving forces. It can be also used for model structure uncertainty different
scenarios for different model structure.

5.6.10.4 Performance According to Some Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.10.3.

• Incorporating qualitative and quantitative information. SC approaches can use
qualitative and quantitative information.
• Automatism. No automatism exists for SC particularly for generation
interdependency among different values of different descriptors. However,
having formalised procedure the process can be organised systematically and
efficiently.
• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: SC approach can have high time and recourses
requirements because of experts or stakeholder involvement.
• Required information: Information needs for quantitative scenarios is on the
level “Int01-04”.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 121

• Required expertise and skills: Some expertise or skills are required for using SC
particularly when using probabilities and Fuzzy inference or clustering.
• Flexibility: The approaches is used commonly in planning phase II. It can be
implemented in different Use Cases.
• Legitimacy: No information is available at this stage of research.

Satisfaction by Planning Participants

• Satisfaction working with method: SC is one of the popular and frequently used
approach.
• Satisfaction by DM with method: In Use Case-II, DM were satisfied with SC
and produced results

5.6.10.5 Future Reading

Except mentioned references above, additional information about scenario

approach can be found in [52–57] or in an extensive review of scenario literature is
done by EEA [58].

5.6.11 Sensitivity Analysis

5.6.11.1 Description

Sensitivity analysis (SA) cited in [10] and defined by Saltelli et al. [59] as “Sensitivity
analysis is the study of how the variation in the output of a model (numerical or
otherwise) can be qualitatively or quantitatively apportioned to different sources of
variation, and of how the outputs of a given model depend upon the information fed
into it”. SA can be distinguished into three different categories [11].
For identification of short list of the most important sensitive factors a screening
SA used to change inputs but not quantifying exactly the total amount of variation
that each factor accounts for [11].
Local SA compute effect of the variation in each input factor keeping others at
some constant level in order to define the rate of change of the output relative to
the rate of change of the input.
Global SA quantifies the effects on the outcomes of interests of variation in the
inputs, as all inputs are allowed to vary over their ranges. The analysis can also
include shapes of probability density function. In this case sampling procedures are
required like Monte Carlo simulation. However, for large models having hundreds of
inputs it is time resources demanding even not possible to have priory information
to specify inputs with their ranges and probability density function for all imputs.
Therefore, screening SA can be very helpful to analysis like Morris method [60].
A so-called one step-at-a-time method is performed giving in each run only one
input parameter a new value. Another method simple approach to define relative
122 5 Multi-method Approaches for Uncertainty Analysis

changes of inputs to outputs is normalized SA screening method suggested by

Morgan et al. [61]. The normalised SA is defined as the ration of relative change of
model output Y induced by a unit relative change in model input X (e.g. percentage).
 
∂y x0
UE (x, y) = ∗
∂x Xo y0

where x0 and y0 are nominal input and output. This measure of uncertainty
importance is also known as elasticity. However, the changes of single inputs are
not contingent on the values that other parameters may take, which is considered
partially by Morris approach.
The results of screening SA can be presented using tornado diagram [62].
The results of global SA are presented using intervals, probability or cumulative
distribution functions.

5.6.11.2 Potential, Main Rationales

It is also used concurrently for triangulation or validation of models or for

development of information.

5.6.11.3 Typology of Uncertainty Addressed

SA typically addresses statistical uncertainty (inexactness) of model inputs and

parameters. It is however also possible to use this technique to analyse sensitivity
of model structure. It provides no insight in the quality of the knowledge base nor
in issues of value loading [11].

5.6.11.4 Performance According to Some Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.11.3.

• Incorporating qualitative and quantitative information. SA approaches use
typically only quantitative information.
• Automatism. High automatism is used in many available software for SA.
• Mixing synergy: See main rationales above.

Organisational Capability

• Required time and recourses: SA do not have high time or recourse requirement.
• Required information: Depending on method the information needs is on the
level “Int02-05”.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 123

• Required expertise and skills: For screening SA no knowledge or additional

skills are required, for global SA using probability information, knowledge
about probabilities and statistics are required.
• Flexibility: The approaches is usually used for planning phase II, it can be
implemented in targeted domain in Use Case-II or in neighbouring domain in
both Use Cases.
• Legitimacy: No information is available at this stage of research.

Satisfaction by Planning Participants

Satisfaction working with method: SA is one of the popular and frequently used
approach.
Satisfaction by DM with method: No information is available at this stage of
research.

5.6.11.5 Future Reading

Compering different SA methods for models with average complexity [63] found that
simple methods like Morris method produce comparable results to those obtained by
more computationally expensive methods. A comprehensive review of SA methods
is presented by Hamby [64]. A guiding book [59] helps non-expert through their
problem in order to enable them to choose and apply the most appropriate SA method.

5.6.12 Tests of Complex Models for Model Uncertainty

5.6.12.1 Description

Every model is wrong. Strong validation of foresight models is not possible

as the modelled real system is not observable in long time horizon from today’s
standpoint. Therefore to have a confidence in developed model structure as
“valid” simplification of real system several model tests have been proposed
and implemented. Barlas [65] propose and describe different tests like tests of
model structure or tests of model behaviour. These tests have been elaborated
and implemented successfully in system dynamics models [66–68] Three major
groups of tests are proposed by Barlas [65] tests of model structure, tests of model
behaviour and tests of policy Implications. They indicate that there is no core set of
tests. However, some tests are frequently used. These tests will be discussed below.
Tests of model structure help to check whether the model structure is an
adequate representation of the real structure or system [69]. Different model
structure tests have been discussed in the literature:
Boundary adequacy (BA): BA test is used for examination conceptual model, to
see whether all problem relevant aspects, system elements, structural relationship
and adequate aggregation level of analysis are considered in the model.
124 5 Multi-method Approaches for Uncertainty Analysis

Structure verification (SV): Domain experts or analysis review of model

assumption and check whether the model structure is consistent with relevant
descriptive knowledge of the system being modelled?
Parameter verification (PV): PV checks the conceptual and numerical
correspondence of model parameters and real system.
Dimensional consistency (DC): DC tests allow analysing the dimensional
consistency of model equations.
Extreme conditions (EC): EC test is useful to check whether the model
behaviour is plausible under certain extreme conditions.
Model behaviour tests are for evaluation of model structure through analysis
of behaviour generated by the structure.
Behaviour reproduction (BR): BR tests how well model generated behaviour
matches observed behaviour of the real system.
Behaviour prediction (BP): Whereas BR tests focus on reproducing historical
behaviour. Two different sub categories exist here. The pattern-prediction test is
to examine whether the model generates qualitatively correct patterns of future
behaviour. The event-prediction test focuses on a particular change in conditions,
such as sharp drop in market share etc. [65].
Behaviour-Anomaly Test (BA): BA can be used e.g. to defend model
assumptions, showing that model behaviour will have anomalous features if
assumption is changed.
Behaviour-sensitivity test (BS): BS checks whether plausible change of model
parameters can cause a model to fail behaviour test previously passed.
“Policy implication tests attempt to verify that response of a real system to a
policy change would correspond to the response predicted by a model. The tests
also examine how robust are policy implications when changes are made in
boundaries or parameters” [65]. Most essential tests as stated are policy-sensitivity
and changed-behaviour prediction tests.
Changed behaviour prediction (CBP): CBP checks whether model correctly
predicts if governing policy is changed.
Policy-sensitivity test (PST): PST reveals the degree to which policy
recommendations might be influenced by uncertainty in parameter values

5.6.12.2 Potential, Main Rationales

The method is usually implemented to develop information for planning or

for other methods. It is also used concurrently for triangulation or validation of
empirical methods.

5.6.12.3 Typology of Uncertainty Addressed

Tests are mainly for model structure, model output or behaviour uncertainty analysis.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 125

5.6.12.4 Performance According to Some Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.12.3.

• Incorporating qualitative and quantitative information. Tests approaches use
both quantitate and qualitative information.
• Automatism. High automatism possible for some tests like behavioural or
extreme conditions tests.
• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: Tests like extreme conditions or dimensional

consistency test have low time and recourses needs. Other tests might need
more time and resources like behavioural or policy conditions test if additional
data or information are required.
• Required information: Information needs is low on the level “Int01-03”.
• Required expertise and skills: For using test no specific knowledge or additional
skills are required.
• Flexibility: Tests are usually used in different planning phase. For example
boundary conditions test can be implemented in planning phase I, other test are
more relevant for planning phase II. Some tests can be implemented in both Use
Cases in different modelling domains.
• Legitimacy: No information is available at this stage of research.

Satisfaction by Planning Participants

• Satisfaction working with method: the implementation of some tests like

boundary or extreme conditions test was satisfactory.
• Satisfaction by DM with method: No information is available at this stage of
research.

5.6.12.5 Future Reading

A detailed discussion about different test have been done by Barlas [65].

5.6.13 NUSAP and PRIMA Methodologies

5.6.13.1 Description

Two overall uncertainty analysis methodologies for integrated assessment models

have been proposed in the last two decades. NUSAP (Numeral Unit Spread
126 5 Multi-method Approaches for Uncertainty Analysis

Assessment Pedigree) as a Delphic ranking procedure approach presented in [70]

and PRIMA (Pluralistic fRamework of Integrated uncertainty Management and
risk Analysis) approach proposed by Van Asselt [71].
Although these approaches have been developed for different problems context,
NUSAP for integrated assessment of anthropogenic climate change and PRIMA
for global change and sustainable development, they share features namely
addressed uncertainty is distinguished into sources and types. Both approaches
support involvement of stakeholder in the analysis taking into account multi
perspective. Whereas the aim of NUSAP is using five qualifiers (Numeral, Unit,
Spread, Assessment and Pedigree) to identify the most uncertain aspect of model
setting priorities to distinguish between the potentially solvable and the currently
unsolvable uncertainties, PRIME incorporate in uncertainty Multi-“perspective”
values paradigms, perceptions of uncertainty based on cultural theory. However, [72]
submit that it oversimplify the reality in cultural theory, being too static in time and
situation (one can be a hierarchist at work or at certain conditions and an egalitarian
at home or in other situation). Both NUSAP and PRIMA are generic frameworks
which can be used with existing techniques and methods. Beside several advantages
like extension of quantitative analysis with qualitative aspects, NUSAP also focus
research effort on the potentially most problematic model property. However, it has
several weaknesses like subjectivity of scoring pedigree criteria and limitation for
implementation on models with large number of parameters [11].

5.6.13.2 Potential, Main Rationales

The method is usually implemented to develop information for planning or

for other methods. It is also used concurrently for triangulation or validation of
empirical methods.

5.6.13.3 Typology of Uncertainty Addressed

Both methodologies can use diverse other methods to address most uncertainty
types mainly for process driven models.

5.6.13.4 Performance According to Quality Factors

Technical Quality Factors

• Extend of incorporation of uncertainty typologies: See discussion in Sect. 5.6.13.3.

• Incorporating qualitative and quantitative information. Both approaches can use
qualitative and quantitative information.
5.6 A Review of Methods and Methodologies for Uncertainty Analysis 127

• Automatism. The methodologies as a hole have low automatism. However,

some sub methods might have certain automatism.
• Mixing synergy: see main rationales above.

Organisational Capability

• Required time and recourses: Both approaches can have very high time and
recourses requirements because of experts or stakeholder involvement.
• Required information: Information needs depends on methods used and can
have low or high data intensity.
• Required expertise and skills: Expertise or skills might be required for some
methods used.
• Flexibility: Both approaches are flexible and can be implemented through all
planning and modelling process. However, for Use Case-II NUSAP approach
want be possible to be implemented fully.
• Legitimacy: Legitimacy of both approaches depends on special city or territory
conditions. No information is available at this stage of research.

Satisfaction by Planning Participants

• Satisfaction working with method: No information is available at this stage of

research.
• Satisfaction by DM with method: No information is available at this stage of
research.

Future Reading

Additional information about NUSAP are in [70] or [73] and for PRIME in [71]
available.

5.7 Summary

Two different multi-method approaches for uncertainty analysis with different

degrees of analytical sophistication and theoretical backgrounds were presented in
this chapter. The initial designs of the methods have been discussed, and potential
methods are indicated, developed and allocated according to the modelling
domain, uncertainty typology, function and role in the multi-method context. Both
approaches will be illustrated in different Use Cases in the next chapter. For the
evaluation of the performance of the FSUA and PRSUA approaches, the quality
factors will be used in Chap. 7.
128 5 Multi-method Approaches for Uncertainty Analysis

References

1. Mirakyan A (2014) PhD thesis: Cadre méthodologique pour la gestion de l’incertain en

planification énergétique long terme des villes et territoires. UNI Strabourg, Strasbourg
2. Paté-Cornell ME (1996) Uncertainties in risk analysis: six levels of treatment. Reliab Eng
Syst Saf 54(2–3):95–111. doi:10.1016/S0951-8320(96)00067-1
3. Mirakyan A, De Guio R (2013) Integrated energy planning in cities and territories: a review
of methods and tools. Renew Sustain Energy Rev 22:289–297. doi:10.1016/j.rser.2013.01.033
4. Mirakyan A, De Guio R (2015) Modelling and uncertainties in integrated energy planning.
Renew Sustain Energy Rev 46:62–69. doi:10.1016/j.rser.2015.02.028
5. Merz B, Thieken AH (2005) Separating natural and epistemic uncertainty in flood frequency
analysis. J Hydrol 309(1–4):114–132. doi:10.1016/j.jhydrol.2004.11.015
6. Hoffman FO, Hammonds JS (1994) Propagation of uncertainty in risk assessments: the
need to distinguish between uncertainty due to lack of knowledge and uncertainty due to
variability. Risk Anal 14(5):707–712. doi:10.1111/j.1539-6924.1994.tb00281.x
7. Hofer E (1996) When to separate uncertainties and when not to separate. Reliab Eng Syst Saf
54(2–3):113–118. doi:10.1016/S0951-8320(96)00068-3
8. Wardekker JA, van der Sluijs JP, Janssen PHM, Kloprogge P, Petersen AC (2008)
Uncertainty communication in environmental assessments: views from the Dutch science-
policy interface. Environ Sci Policy 11(7):627–641. doi:10.1016/j.envsci.2008.05.005
9. Kloprogge P, Sluijs Jvd, Wardekker A (2007) Uncertainty communication, issues and good
practice
10. Refsgaard JC, van der Sluijs JP, Højberg AL, Vanrolleghem PA (2007) Uncertainty in the
environmental modelling process—a framework and guidance. Environ Model Softw
22(11):1543–1556. doi:10.1016/j.envsoft.2007.02.004
11. Van der Sluijs JP, Janssen PHM, Petersen AC, Kloprogge P, Risbey JS, Tuinstra W, Ravetz JR
(2004) RIVM/MNP guidance for uncertainty assessment and communication: tool catalogue
for uncertainty assessment (trans: Innovation CIfSDa). BA Bilthoven
12. Ross TJ, Booker JM, Parkinson WJ (2002) Fuzzy logic and probability applications: bridging
the gap. Society for Industrial and Applied Mathematics (SIAM, 3600 Market Street, Floor 6,
Philadelphia, PA 19104)
13. Mirakyan A, De Guio R (2014) A methodology in innovative support of the integrated
energy planning preparation and orientation phase. Energy 78:916–927. doi:10.1016/j.
energy.2014.10.089
14. Lawrence M, Goodwin P, O’Connor M, Önkal D (2006) Judgmental forecasting: a review of
progress over the last 25 years. Int J Forecast 22(3):493–518. doi:10.1016/j.ijforecast.2006.03.007
15. Sterman J (2000) Business dynamics: systems thinking and modeling for a complex world.
Irwin/McGraw-Hill c2000
16. Bukowski J, Korn L, Wartenberg D (1995) Correlated inputs in quantitative risk assessment:
the effects of distributional shape. Risk Anal 15(2):215–219. doi:10.1111/j.1539-6924.1995.
tb00315.x
17. Jaworski P, Durante F, Härdle WK, Rychlik T (2010) Copula theory and its applications. In:
Proceedings of the workshop held in Warsaw, 25–26 Sept 2009. Springer, Berlin
18. Buckley JJ (1985) Entropy principles in decision making under risk. Risk Anal 5(4):303–
313. doi:10.1111/j.1539-6924.1985.tb00186.x
19. Clemen RT, Reilly T (1999) Correlations and copulas for decision and risk analysis. Manage
Sci 45(2):208–224. doi:10.1287/mnsc.45.2.208
20. Abbas AE (2006) Entropy Methods for Joint Distributions in Decision Analysis. IEEE
Transactions on engineering managment 53 (1)
21. Nelsen RB (2006) An introduction to copulas. Springer Science & Business Media
22. Knol A, Slottje P, van der Sluijs J, Lebret E (2010) The use of expert elicitation in
environmental health impact assessment: a seven step procedure. Environ Health 9(1):19
References 129

23. Spetzler CS, Carl-Axel SSVH (1975) Probability encoding in decision analysis. Manage Sci
22(3):340–358. doi:10.2307/2629647
24. O’Hagan A (2012) Probabilistic uncertainty specification: overview, elaboration techniques
and their application to a mechanistic model of carbon flux. Environ Model Softw 36:35–48.
doi:10.1016/j.envsoft.2011.03.003
25. Booker JM, Anderson MC, Meyer MA (2001) The role of expert knowledge in uncertainty
quantification (Are we adding more uncertainty or more understanding?), pp 155–161.
doi:citeulike-article-id:8495306
26. Tversky A, Kahneman D (1974) Judgment under uncertainty: heuristics and biases. Science
185(4157):1124–1131. doi:10.2307/1738360
27. Mamdani EH, Assilian S (1975) An experiment in linguistic synthesis with a fuzzy logic
controller. Int J Man Mach Stud 7(1):1–13. doi:10.1016/S0020-7373(75)80002-2
28. Chen G, Pham TT (2000) Introduction to fuzzy sets, fuzzy logic, and fuzzy control systems.
Taylor & Francis
29. Siler W, Buckley JJ (2005) Fuzzy expert systems and fuzzy reasoning. Wiley-Interscience
30. Celikyilmaz A, Türksen IB (2009) Modeling uncertainty with fuzzy logic: with recent theory
and applications. Springer, Heidelberg
31. Zimmermann HJ (2001) Fuzzy set theory-and its applications. Kluwer Academic Publishers
32. Lawry J, Miranda E, Bugarin A, Li S, Gil MA, Grzegorzewski P, Hryniewicz O (2007) Soft
methods for integrated uncertainty modelling
33. Hyndman RJ, Khandakar Y (2008) Automatic time series forecasting: the forecast package
for R. J Stat Softw 27(3):1–22
34. Chatfield C (1993) Calculating interval forecasts. J Bus Econ Stat 11(2):121–135.
doi:10.2307/1391361
35. Chatfield C (2002) Time-series forecasting. Taylor & Francis
36. Williams WH, Goodman ML (1971) A simple method for the construction of
empirical confidence limits for economic forecasts. J Am Stat Assoc 66(336):752–754.
doi:10.2307/2284223
37. Wright G, Lawrence MJ, Collopy F (1996) The role and validity of judgment in forecasting.
Int J Forecast 12(1):1–8. doi:10.1016/0169-2070(96)00674-7
38. Everette S, Gardner J (1988) A simple method of computing prediction intervals for time
series forecasts. Manage Sci 34(4):541–546. doi:10.1287/mnsc.34.4.541
39. Hyndman RJ, Koehler AB, Ord JK, Snyder RD (2005) Prediction intervals for exponential
smoothing using two new classes of state space models. J Forecast 24(1):17–37. doi:10.1002/
for.938
40. Snyder RD, Ord JK, Koehler AB (2001) Prediction intervals for ARIMA models. J Bus Econ
Stat 19(2):217–225. doi:10.2307/1392165
41. Fishman GS (2006) A first course in Monte Carlo. Thomson Brooks/Cole, Duxbury
42. Refsgaard JC, van der Sluijs JP, Brown J, van der Keur P (2006) A framework for dealing
with uncertainty due to model structure error. Adv Water Resour 29(11):1586–1597.
doi:10.1016/j.advwatres.2005.11.013
43. Visser H, Folkert RJM, Hoekstra J, de Wolff JJ (2000) Identifying key sources of uncertainty in
climate change projections. Clim Change 45(3–4):421–457. doi:10.1023/a:1005516020996
44. Hyndman RJ, Koehler AB (2006) Another look at measures of forecast accuracy. Int J
Forecast 22(4):679–688. doi:10.1016/j.ijforecast.2006.03.001
45. Armstrong JS, Collopy F (1993) Error measures for generalizing about forecasting methods:
empirical comparisons. Int J Forecast 8(1) 69–80. doi:http://dx.doi.org/10.1016/0024-6301(93)
90280-S (June 1992. Long Range Planning 26(1):150 )
46. Carbone R, Armstrong JS (1982) Evaluation of extrapolative forecasting methods: results of a
survey of academicians and practitioners. J Forecast 1(2):215–217. doi:10.1002/for.3980010207
47. Zürni SU (2003) Möglichkeiten und Grenzen der Methode der Szenarioanalyse in der
Öffentlichen Planung: eine Analyse am Beispiel der Schweizer Energieplanung wiku-verlag
48. Mißler-Behr M (1993) Methoden der Szenarioanalyse. Deutscher Universitäts-Verlag, Univ,
Augsburg
130 5 Multi-method Approaches for Uncertainty Analysis

49. Götze U (1991) Szenario-Technik in der strategischen Unternehmungsplanung. DUV:

Wirtschaftswissenschaft edn. Deutscher Universitäts-Verlag
50. Mißler-Behr M (2001) Fuzzybasierte Controllinginstrumente. Entwicklung von unscharfen
Ansätzen. vol 1. Deutscher Universitäts-Verlag
51. Bezdek JC (1981) Pattern recognition with fuzzy objective function algorithms. Kluwer
Academic Publishers
52. Schwartz P (1996) The art of the long view: paths to strategic insight for yourself and your
company. Currency Doubleday
53. van der Heijden K (2011) Scenarios: the art of strategic conversation. Wiley
54. Sumerling T (2001) Scenario development methods and practice: an evaluation based on the
NEA workshop on scenario development, Madrid, May 1999. OECD
55. Schnaars SP (1987) How to develop and use scenarios. Long Range Plan 20(1):105–114.
doi:10.1016/0024-6301(87)90038-0
56. Swart RJ, Raskin P, Robinson J (2004) The problem of the future: sustainability science and
scenario analysis. Glob Environ Change 14(2):137–146. doi:10.1016/j.gloenvcha.2003.10.002
57. Börjeson L, Höjer M, Dreborg K-H, Ekvall T, Finnveden G (2006) Scenario types and
techniques: towards a user’s guide. Futures 38(7):723–739. doi:10.1016/j.futures.2005.12.002
58. European Environment Agency (EEA) (2009) Looking back on looking forward: a review of
evaluative scenario literature, vol No 3/2009. Technical report. Copenhagen
59. Saltelli A, Tarantola S, Campolongo F, Ratto M (2004) Sensitivity analysis in practice: a
guide to assessing scientific models. Wiley
60. Morris MD (1991) Factorial sampling plans for preliminary computational experiments.
Technometrics 33(2):161–174. doi:10.2307/1269043
61. Morgan MG, Henrion M, Small M (1992) Uncertainty: a guide to dealing with uncertainty in
quantitative risk and policy analysis. Cambridge University Press, Cambridge
62. Clemen RT (1996) Making hard decisions: an introduction to decision analysis. Duxbury
Press
63. Confalonieri R, Bellocchi G, Bregaglio S, Donatelli M, Acutis M (2010) Comparison
of sensitivity analysis techniques: a case study with the rice model WARM. Ecol Model
221(16):1897–1906. doi:10.1016/j.ecolmodel.2010.04.021
64. Hamby DM (1994) A review of techniques for parameter sensitivity analysis of
environmental models. Environ Monit Assess 32(2):135–154. doi:10.1007/bf00547132
65. Forrester JW, Senge PM (1978) Tests for building confidence in system dynamics models.
System Dynamics Group, Sloan School of Management, Massachusetts Institute of Technology
66. Barlas Y (1994) Model validation in system dynamics. Paper presented at the proceedings of
the 1994 international system dynamics conference, Scotland
67. Barlas Y, Carpenter S (1990) Philosophical roots of model validation: two paradigms. Syst
Dyn Rev 6(2):148–166. doi:10.1002/sdr.4260060203
68. Qudrat-Ullah H, Seong BS (2010) How to do structural validity of a system dynamics type
simulation model: the case of an energy policy model. Energy Policy 38(5):2216–2224.
doi:10.1016/j.enpol.2009.12.009
69. Barlas Y (1989) Multiple tests for validation of system dynamics type of simulation models.
Eur J Oper Res 42(1):59–87. doi:10.1016/0377-2217(89)90059-3
70. Funtowicz SO, Ravetz JR (1990) Uncertainty and quality in science for policy. Springer,
Berlin
71. Van Asselt BA (2000) Perspectives on uncertainty and risk: the PRIMA approach to decision
support. Springer, Berlin
72. Trisoglio A (1995) Plurality and political culture, a response to thompson and drake. In: Risk,
policy and complexity, Laxenburg, IIASA, p 18
73. Van Der Sluijs JP, Craye M, Funtowicz S, Kloprogge P, Ravetz J, Risbey J (2005) Combining
quantitative and qualitative measures of uncertainty in model-based environmental assessment:
the NUSAP system. Risk Anal 25(2):481–492. doi:10.1111/j.1539-6924.2005.00604.x
Chapter 6
Implementation of Discussed Uncertainty
Analysis Approaches in Case Studies

The objectives of this chapter are

• Presentation of the implementation of the 3-domain modelling notion in differ-
ent Use Cases and discussing some of the results
• Demonstrating the implementation of multi-method approaches for uncertainty
analysis: FSUA and PRSUA in different Use Cases
• Discussing the return of the experience and results generated by the modelling
and uncertainty analysis approaches

6.1 Selection of Application Studies

Two different types of Use Cases in the context of IEPCT have been identified in
Chap. 2, Sect. 2.4. Similar types of planning conditions appear in the cooperation
projects at the EIFER institute. In one study related to Use Case I—Singapore,
there was an overall vision for “Singapore sustainable growth”. However, because
of the starting conditions of the study, it was not possible to develop a single, inte-
grated, quantitative model at the beginning. This type of situation is not unique, as
experience shows that it is not easy to start to develop an integrated multi-sector
model because of various difficulties, e.g., missing data or expertise availability,
different sector departments using different approaches, and terminologies or plan-
ning rationales. Nevertheless, even in these situations, an overall view of the city
development is required, which links the different single-sector models together, at
least in a qualitative manner. In the second study, related to Use Case II—Mexico
City, there was a clear need to develop a single, quantitative integrated “waste to
energy” model for Mexico City at the beginning of the study.

© Springer International Publishing Switzerland 2015 131

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_6
132 6 Implementation of Discussed Uncertainty Analysis Approaches …

Presented uncertainty analysis approaches are implemented in Use Cases.

FSUA can be implemented in both Use Cases, whereas PSRUA cannot be imple-
mented for Use Case I—Singapore, where there is no available integrated quan-
titative model developed. However, some single methods of both approaches
are applicable in both Use Cases. The application of the FSUA approach will be
demonstrated and evaluated in Sects. 6.2 and 7.2.2 respectively using Use Case
I—Singapore. The application of the PRSUA approach will be demonstrated and
evaluated in Sects. 6.3 and 7.2.3 respectively using Use Case II—Mexico City.

6.2 An Example of Use Case I: Singapore

The main study objective was the development of a plan for sustainable growth
in Singapore. Different models, such as a cooling system or lighting system, were
planned to be developed. However, it was only possible to integrate different mod-
els in a qualitative way at the beginning of planning. An overall uncertainty analy-
sis was required for all modelling domains addressing all types of uncertainties.
The data implemented in modelling and uncertainty analysis are publicly available
data, which had been found in journals, books or in internet.
At the beginning, the Singapore sustainable growth model was developed. The
modelling was performed according to the 3-domain modelling concept presented in
Chap. 3. Firstly defined modelling methods had be applied to each domain. In par-
allel, data were collected for uncertainty analysis, which was performed during the
modelling process. Secondly, uncertainty analysis in different domains in performed.
The performance of the applied methods was evaluated after implementation.

6.2.1 Development of the “Singapore Sustainable

Growth” Model

6.2.1.1 Historical and Current Situation

Singapore is an island city-state located in South-East Asia. It has a total land

area of 699.4 km2 and a population of ca. 5.18 million as of 2011. The popula-
tion increased from approximately 4.02 million in 2000, which corresponds to an
annual average growth rate of 2.6 %. It is expected that, in the future, the popula-
tion growth will be lower than it has been in the past 15 years.
Although Singapore is small in terms of land area and population, its economy
has seen rapid growth, making the country one of the most industrialised and
urbanised in South-East Asia. The Gross domestic product (GDP) per capita grew
from 29.4 thousand USD/person in 2005 to 43.87 thousand USD/person in 2010.
For the period of 2010–2030, it is expected that, as Singapore’s economy matures,
the GDP will grow at a slower rate compared to the period of 2005–2030 [1].
6.2 An Example of Use Case I: Singapore 133

Singapore has no sufficient indigenous energy resources. Singapore lacks the

natural endowments to tap hydropower or geothermal energy. Wind speed limits
the use of wind power with current technology. Only biofuel, solar energy and
waste-to-energy would be applicable. However, Singapore is a major oil-refin-
ing centre for the Southeast Asian region and is a major exporter of petroleum
products. Singapore imports natural gas from Malaysia and Indonesia for power
generation. By 2012, Singapore had a LNG terminal with the initial capacity of
3 million tons per annum serving Singapore’s rising demands for energy.
The total final energy consumption of Singapore has grown at an average
annual growth rate of 5.7 %, from approximately 6839 Kilo Tone Oil Equivalent
(KTOE) in 1990 to 15616 KTOE in 2005. The total final energy consumption
in the industrial sector, which consumed the largest portion of energy, grew at
an average rate of 6.9 % per annum. Energy consumption in the domestic sec-
tor, although only responsible for approximately 5.9 % of the total energy con-
sumption in 2005, grew at an average annual rate of 5.5 % from 1990 to 2005.
The commercial sector energy consumption grew at an average rate of 7.1 %
per annum. The total energy consumption in the transport sector saw an average
annual rate increase of 3.8 % for the period from 1990 to 2005 [1, 2].

6.2.1.2 Modelling the Targeted Domain and Identification

of Key Descriptors

The targeted domain is the entire city, which considers different sub models, such
as the energy generation model, lighting model, cooling model, etc. However,
it was necessary to have an integrated view of all of these required models.
Moreover, individual models were not complete at that stage of the planning. For
this type of situation, particularly when there is little data available at the beginning
and certain expert knowledge is available, judgment driven models can be helpful.
The MICMAC approach, discussed in section “MICMAC Approach to Model the
Targeted Zone in Case 1—Singapore”, has been implemented to develop the qualita-
tive model of the study area, taking into account the descriptors of all sub models.
The picture on the top left side of Fig. 6.1 shows the interaction among d­ ifferent
city descriptors. The picture on the right side shows the output of analysis, i.e., the
allocation of different city descriptors according to influence and dependency. The
descriptors are mostly allocated along the dependency or influence axes, form-
ing similar as ‘L’ shape, which indicates stability of the overall model and low
dependency of different sub models, such as the cooling system or solar PV model,
on each other.
After collecting all the inputs of the sub models and including some general
descriptors of the city, such as gross domestic product (GDP), electricity tar-
iffs (Elec-tarif) etc. interdependency among these model inputs and city general
descriptors was established using qualitative terms of the MICMAC approach. The
analysis of the interaction among descriptors was performed in three steps. In the
first step, the analysis of the direct influence of descriptors is conducted; in the
 134
6

Fig. 6.1  Qualitative “Singapore sustainable growth” model and identification of key descriptors
Implementation of Discussed Uncertainty Analysis Approaches …
6.2 An Example of Use Case I: Singapore 135

Fig. 6.2  Evolution of descriptor importance ranking according to influence

second step, the indirect cumulated influence is analysed; and in the third step, the
potential future influence is integrated into the analysis.
Descriptors in the highlighted area in the top left rectangle of Fig. 6.1 are those
that have high influence and low dependence. These descriptors have to be ana-
lysed in more detail and included in scenario building for integrated city model in
subsequent steps.
Figure 6.2 shows the ranking of different descriptors according to their impor-
tance (having high influence on other descriptors and low dependency). The first
column shows the ranking when only direct interdependency among different
descriptors is considered. The second column shows the ranking of descriptors
when both direct and non-direct influences are included. The third column, on the
right side, shows the descriptor order when potential influence of the descriptors,
which might have importance in the future, are also considered. Green and red
lines link a descriptor’s different ranks, which vary as information about influence
increases or decreases. A green line indicates that the importance of the descriptor
increases when indirect or potential influence among descriptors are considered. A
red line indicates the opposite.
136 6 Implementation of Discussed Uncertainty Analysis Approaches …

The analysis shows significant changes to the level of influence of descriptors

when indirect influence is included. There are also some moves in the order of
influence based on the long-range third column when future potential influences
are included. For example Gas-tariff descriptor (variable) had an influence rank
27. However, considering indirect influence (multiplying the matrix many times)
Gas-tariff descriptors becomes the influence rank 12 and so one.
As was discussed in section “MICMAC Approach to Model the Targeted Zone
in Case 1—Singapore”, in Chap. 3 the highlighted area (see Fig. 6.1) is the area
of primary importance for long-range scenario generation and strategic planning.
Descriptors with low dependence and influence and descriptors with high depend-
ence but low influence are excluded from the future analysis, as they are not con-
sidered significant for the overall long-term scenario development. Descriptors
with high influence and dependence are not considered in scenarios either because
of their high interrelation with other descriptors, potential instability and low pre-
dictability. Table 6.1 shows the final list of descriptors extracted from the analysis
Fig. 6.1, enumerated according to their importance.
For the next step of analysis, 15 descriptors having highest influence out of the
total 45, have been identified.

Table 6.1  Final list of key descriptors

N° Long label xx-Short Description Theme
label
1 GDP GDP S$ mil/a Economic
2 Demography Demog. % Sociodemographic
chang_share Share
3 Average monthly HH income S$/month Economic
household income
4 Household size HH/size HH/person Sociodemographic
5 Life style LifeStyle More or less comfort Sociodemographic
oriented
6 Electricity tariffs Elec_tarif S$/kWh Economic institutional
7 Daily solar radiation Solar kWh/m2 Environment-climate
8 Population Pop Persons Sociodemographic
9 Building types BuildTyp Category Technological
10 Yearly electricity ElecCon/HH MWh/(a * HH) Sociodemographic
consumption per economic
household
11 Gas tariffs Gas_tarif S$/m3 Economic institutional
12 Outdoor temperature Temp °C Environment-climate
13 New regulations ReglRestr Restriction of some Institutional
environmental pollutants or
restrictions efficiency regul.
14 Discount rate DiscRate % Economic
15 Diffusion of new eff. DiffAirCon %/(HH * a) Socioeconomic
air condit system
6.2 An Example of Use Case I: Singapore 137

Fig. 6.3  Population growth in Singapore multi model simulation, composite extrapolation

(Data, historical data; Theta, arima (0.2.2), arima (1.2.2) is generated via auto.arima; nnetar,
neural networks; RandWalkFwD, random walk with drift or ‘trend’; LinReg, Linear Regression,
RobTrend, Robust trend [see more details in Sect. 3.4.3].)

6.2.1.3 Modelling the Neighbouring Domain

The analysis of the neighbouring domain is concentrated now on 15 key descrip-

tors, which are the most important for the targeted domain identified before. For
modelling key descriptors, pre-selected data driven and judgmental methods, iden-
tified in Sect. 3.4.3, have been implemented. Different data analysis techniques
exist for historical data analysis.
The focus of this book is not the descriptive but inductive analysis. However,
some results of descriptive analysis like decomposition of time series, identifica-
tion of outliners are presented in Appendix A.
Using preselected data driven modelling approaches inductive analysis of key
descriptors values have been performed (Fig. 6.3).
After comparing the methods according to MASE and MdAPE value (see
Sect. 5.6.9 in Chap. 5) data driven methods are compared. However, the orienta-
tion is not only to select the best approach according to MASE or MdAPE value
alone but with composite extrapolation make triangulation of results. The multi
method simulation using different models and making composite extrapolation is
a way to deal with model structure uncertainty. Some selected results of inductive
analysis are presented in Appendix B. The final quality of certain method or model
depends also on generated drift which will be discussed in section “Model Inputs
Uncertainty Analysis in the Targeted Domain”.
138 6 Implementation of Discussed Uncertainty Analysis Approaches …

All pre-selected data driven models have been implemented for more than 29
datasets regarding the demography, economy, etc. of Singapore. The performance
of methods across all data sets according to quality criteria is discussed in section
“Model Structure uncertainty in the Neighbouring Domain”.

6.2.1.4 Modelling the Distant Domain

A reference impact matrix (RIM) was used to model the distant domain and
incorporate information for the neighbouring domain and targeted domain. Using
various categories, the current and possible future situations are presented in
Table 6.2, based on a literature review and expert judgment. The theoretical back-
ground or RIM is explained in Sect. 3.4.4.1.
An example to clarify the output of the matrix key descriptor ‘temperature’
in Singapore is considered. According to the weak signals WS8, WS12 and WS
13 the temperature in model context will have some transient effects. However,
the trend of increasing temperature will stay. Extend of impact is evaluated and
discussed in section “Model Inputs Uncertainty Analysis of the Neighbouring
Domain”.

6.2.2 Uncertainty Analysis

6.2.2.1 Context and Framing Uncertainty Analysis

A model context uncertainty analysis was performed using IMMA [6]. The devel-
opment and framing of the conceptual model, and uncertainty analysis of model
framing using the conformity checklist for quality assurance are demonstrated in
[6]. Also uncertainty related to ambiguity about objectives and initiatives are dis-
cussed in this study.

6.2.2.2 Model Structure Uncertainty Analysis

Model Structure Uncertainty in the Targeted Domain

The developed qualitative model for the targeted domain is based on qualitative
interrelation among city descriptors. The uncertainties of interdependency are
incorporated using fuzzy linguistic terms and expert elicitation according to the
MICMAC1 approach. Additionally, judgmental sensitivity analysis is used to see
whether the position and influence ranks of descriptors change. There were some
change of key descriptors position or rankings see Figs. 6.1 and 6.2.

1Matrice d’Impacts Croiés—Multiplication Appliquée à un Classement.

Table 6.2  Reference impact matrix for Singapore
Impact category Trends-cycles (T) Discontinuities (D) Weak signals (WS) Wild cards (WdC)
Frame of reference
Global- 2008–2015 2008–2015 2008–2015 2008–2015
international [3, 4]
T1. Towards Miniature D1. Communication WS1. Recurrent wars and conflicts WdC1. Wars and Conflicts
Mobile Technologies Technologies WS2. Economic Recession WdC2. Natural Disasters
T2. Environmental- D2. Shift to Alternative WS3. Enhancement of Innovation WdC3. Decline of World
Sustainability Awareness Energy Economy WS4. Rising Individualism and Order and Collapse of
T3. Changing Socio- D3. Socio-Political Crisis Nationalism Nations
Economic Patterns D4. Scarcity of Resources WS5. Lack of Education and WdC4. Terrorism
T4. Increasing Conflicts D5. Terrorism and Security Development Funds WdC5. Increasing Migration
T5. Enhancement of Threats Flows
6.2 An Example of Use Case I: Singapore

Science and Innovation

2016–2025 2016–2025 2016–2025 2016–2025
T6. Environmental- D6. Shift to Alternative WS6. Recurrent wars and conflicts WdC6. Epidemic Outbreaks
Sustainability Awareness Energy Economy WS7. Emerging Societies WdC7. Decline of World
T7. Alternative Energy D7. Changing Political WS8. Consequences of Climate Change Order and Collapse of
Sources Systems WS9. Artificial Intelligence Applications Nations
T8. Enhancement of D8. Sustainable Life Styles WS10. New World Order and Patterns of WdC8. Wars and Conflicts
Science and Innovation D9. Wars and Conflicts Democracy WdC9. Natural Disasters
T9. Increased Mobility D10. Precautionary Principle WdC10. Terrorism
and Migration flows and Ethical Values
T10. Towards Miniature
Mobile Technologies
(continued)
139
Table 6.2  (continued)
140

Impact category Trends-cycles (T) Discontinuities (D) Weak signals (WS) Wild cards (WdC)
Frame of reference
Beyond 2025 Beyond 2025 Beyond 2025 Beyond 2025
T11. Environmental- D11. Shift to Alternative WS11. Artificial Intelligence Applications WdC11. Natural Disasters
Sustainability Awareness Energy Scarcity of WS12. Consequences of Climate Change WdC12. Technological
T12. Scarce Natural Resources WS13. Efficient Energy Policies Breakthroughs
Resources D12. Climate Change WS14. New World Order and Patterns of WdC13. Energy and Oil Crisis
6

T13. Ageing Population D13. Sustainable Life Styles Democracy WdC14. New Political/
T14. Alternative Energy D14. New Advances in WS15. Availability of Information and Economic Alliances
Sources Medicine Confidentiality WdC15. Wars and Conflicts
Transnational D15. Smog or haze from fire
in Indonesia
National [5] 1990–2005 1990–2005 –2013 2008–2015
T15. Very high value D16. New information and WS16. Increasing temperature WdC16. From WC5
added technology & communication technologies Increasing Migration Flows
knowledge based manu-
facturing and services
City, territory, See above national level See above national level See above national level See above national level
neighbourhood
Impacted model 2008–2015 2008–2015 2008– 2008–2015
domain, feature,
variables, param-
eters etc.
T16. Energy consump- WS17. Increasing temperature (WS8, WdC17. Demography.
tion reduction (T1;T2) WS12, WS 13) Affected by (WdC5)
T17. Increase technology WdC18. Energy and fuel
efficiency (T5) prices (WdC1;2;3)
(continued)
Implementation of Discussed Uncertainty Analysis Approaches …
Table 6.2  (continued)
Impact category Trends-cycles (T) Discontinuities (D) Weak signals (WS) Wild cards (WdC)
Frame of reference
2016–2025 2016–2025 2015–
T18. Energy consump- D17. Chang of consumption WdC19. Energy supply
tion reduction (T6) patterns (D1) and (D8) security (WdC1; 2; 3; 6, 8,10)
T19. Increase technology D18. High diffusion rate of
efficiency (T8) photovoltaic panels (D6)
D19. Change of Temperature,
Solar radiation (D15)
2016–2025 2016–2025 2025–
T20. Change of D20. Chang of consumption WdC19. Energy supply
Temperature, Solar radia- patterns (D8) security (WC11, 13, 15)
6.2 An Example of Use Case I: Singapore

tion (T11) D21. Change of

T21. Increase of energy Temperature, Solar radiation
price (T2; 4) (T11)
T22. Intermittency of
energy supply (T11, T13,
T15)
T23. Demography and
consumption pattern
changes (T7, T14)
Type of impacta T16, 18. Trend change D17. Trend change WS17. Trend and transient change WdC17. Quantum jump
T17, 19. Trend change D18. Transient change WdC18. Quantum jump
for technologies D19. Transient change WdC19. Transient change
efficiency D20. Transfer change
T20. Trend change D21. Trend & transient
T21. Trend, transfer change
change
T22. Transient change
T23. Trend change
141

aSee discussion about type of impact in Sect. 3.4.3

142 6 Implementation of Discussed Uncertainty Analysis Approaches …

Table 6.3  Average performance of data driven models and methods across all data sets for some
data of Singapore
Methods, Theta ARIMA auto.ARIMA Nnet LinReg Random RobTrend
models (0.2.2) walk with
drift
MASE 0.943 0.493 0.509 0.706 2.146 0.815 0.808
MdAPE 0.041 0.035 0.037 0.035 0.070 0.034 0.034
Theta model, arima (0.2.2) model, arima (1.2.2) is generated via auto.arima; nnetar, neu-
ral networks; RandWalkFwD, random walk with drift or ‘trend’; LinReg, Linear Regression,
RobTrend, Robust trend (see more details in Sect. 3.4.3)

Model Structure uncertainty in the Neighbouring Domain

Data driven model structure uncertainty has been performed using the multi model
simulation MMS method (see Sect. 5.6.9). Pre-defined methods and the models
from Sect. 3.4.3.2 have been used for the same data set to identify not only the
most appropriate method according to the MASE and MdAPE values but also to
make triangulation. The performances of methods are very different for long-term
extrapolation and depend on the available historical data set.
The analysis (see Table 6.3) shows that, for large amount of memory data or
available observation (more than 12 historical periods), for 29 data sets from dif-
ferent domains, including demography and economy, the best MASE value has
ARIMA (0.2.2), automatic ARIMA and neural networks. The least best MdAPE
values show Robust Trend, random walk and ARIMA (0.2.2) and neural networks.
There is no clear ranking among different methods when considering MASE
and MdAPE jointly. The performance of tested methods and models depends
strongly on the available data amount and character of the individual data set. The
Use Case—II Mexico city will help to clarify ranking using additional data sets.
However, less than 12 memory data were not enough for the neuronal nets method
for recognition of plausible patterns. Neuronal nets showed better results when
there were large amounts (more than 20 historical periods) of memory data avail-
able especially when using quartile or monthly data. More results of the analysis
are in Appendix B.
Using multi model simulation MMS for triangulation gives the opportunity to
analyse, compare different models result and make composite extrapolation and
model structure uncertainty analysis. Ideally, triangulated results by different
methods have to be in insight the drift generated by best fitted method or model.

Model Structure Uncertainty in the Distant Domain

The distant domain is modelled using a reference impact matrix and expert elic-
itation. Uncertainty is mainly related to fixed terminologies and categories such
as trend, discontinuities, etc. used in RIM. These categories have been already
6.2 An Example of Use Case I: Singapore 143

discussed and evaluated in large expert community see Sect. 3.3.3.1, therefore no
uncertainty analysis of the model structure of the distant domain is explicitly per-
formed additionally in this study.

6.2.2.3 Model Inputs Uncertainty Analysis

Model Inputs Uncertainty Analysis in the Targeted Domain

Uncertainty of the model inputs exists for each single driving force, as well as
because of the interaction among different forces.

Uncertainty of Individual Model Driving Forces

Model input uncertainty is presented using prediction intervals. Different interval
prediction approaches (see Sect. 5.6.6) have been implemented to define the inter-
vals. The accuracy and precision of different intervals depends on the model and
method used for extrapolation, the character of the dataset and the interval analysis
method. Model inputs uncertainty analysis of process driven models is model out-
put uncertainty analysis of data driven models (Fig. 6.4).

Fig. 6.4  Uncertainty of model driving forces, different intervals (Normal, based on normality
assumption of residue; Chebychev, interval method based on Chebychev’s inequality, see more
details in Sect. 5.6.6)
144 6 Implementation of Discussed Uncertainty Analysis Approaches …

The analysis shows that intervals generated by ‘normal’ interval approach is

relative narrow across all data sets, intervals approach based on Chebychev’s ine-
quality generate very large intervals and is not stable when there are few histori-
cal data available. However, the normal distribution of residue has to be checked
before using ‘normal’ interval approach. 43 % of all available data sets show
­normal distribution of residue after fitting the model.

Uncertainty Because of Interdependency Among Different Model Inputs

and Linguistic Uncertainty
In this section, interdependency among driving forces of the targeted domain is
discussed. Szeno-Plan software [7] is used for implementation and evaluation of
this interdependency. Two approaches are implemented for analysis interdepend-
ency among driving forces: consistency analysis and cross-impact analysis with
fuzzy inference for non-precise probability estimation (see the theoretical discus-
sion in methods scenario in Sect. 5.6.10).
The consistency matrix for the analysis of key factors values is shown in
Fig. 6.5.
As an example, the consistency among two descriptors, household energy con-
sumption and household income, is defined as follows: High household income
and high household energy consumption are strong consistent = 2; or 90 quintile
of electricity tariff growth have weak inconsistency with 25 quintile of gas tariff
growth = −1.

Fig. 6.5  Consistency analysis of the targeted domain key descriptors development of Singapore
6.2 An Example of Use Case I: Singapore 145

Fig. 6.6  Cross-impact analysis of the targeted domain key descriptors development of Singapore

The cross-impact (CIM) analysis helps to define posterior probabilities of dif-

ferent input key factors values using their a priori probability and identify scenar-
ios that have the highest appearance occurrences in the total set (Fig. 6.6).
Cross-impact analyses have been performed using fuzzy inference to reduce
linguistic uncertainty in expert elicitation and incorporate non precise information.
The trapezoidal membership function is used, as it is very common (Fig. 6.7).
Linguistic terms for the antecedent: CIM conditional probabilities are
Y = (“extreme reduction”, ‘strong reduction’, ‘weak reduction’, ‘neutral’,
‘weak growing’, ‘strong growing’, ‘extreme growing’)
Linguistic terms for the antecedent: a priori probability and for consequent pos-
terior probability are
X = (‘very seldom’, ‘seldom’, ‘normal’, ‘often’, ‘very often’)
If a priori probability P = ‘seldom’ and value in C-I-Matrix = ‘strong reduc-
tion’ then posterior probability P = ‘very seldom’.
For example, if a priori probability of population growth is ‘normal’, which is
in the area of between −25 and +25 quintile of the interval, and the migration
effect has ‘extreme growing’ impact, then posteriorly probability of population
growth will be very often, which corresponds to a value of 95 quintile of the popu-
lation growth (Table 6.4).
Mamdani a rule-inference method is implemented (see Table 6.5).
146 6 Implementation of Discussed Uncertainty Analysis Approaches …

Fig. 6.7  Trapezoidal membership functions implemented in the study

Table 6.4  Linguistic terms for cross-impact value

Cross- −3 −2 −1 0 1 2 3
impact
value
Linguistic Extreme Strong Weak Neutral Weak Strong Extreme
terms reduction reduction reduction growing growing growing

Table 6.5  An example of structure of the Mamdani-style inference implemented

If-then Extreme Strong Weak Neutral Weak Strong Extreme
rules reduction reduction reduction growing growing growing
Very Very Very Very Very Seldom Often Often
seldom seldom seldom seldom seldom
Seldom Very Very Very Seldom Normal Often Often
seldom seldom seldom
Normal Seldom Seldom Seldom Normal Normal Very Very often
often
Often Seldom Normal Normal Often Often Very Very often
often
Very Seldom Normal Often Very Very Very Very often
often often often often

Model Inputs Uncertainty Analysis of the Neighbouring Domain

Despite the randomness of model inputs, which is presented using intervals (see
section “Uncertainty of Individual Model Driving Forces”), uncertainty from distant
domain are incorporated using causal forces (see Sect. 3.3.3.1). Reference impact
6.2 An Example of Use Case I: Singapore 147

matrix (RIM) helps to identify the model inputs and type of impact coming from
distant domain. Different approaches have been presented how to incorporate the
extent of impact in Sect. 3.3.3.1. The relevancy of a causal force for neighbouring
domain is evaluated in this case study by following overall question: Which data
driven method or model with a certain interval has the best MASE and MdAPE
value, as well as an extrapolated mean value that is still in the generated interval
if the impact of causal force is equal to the highest residue for all types of impacts
discussed in Sect. 3.3.3.1? The focus was selected data driven methods and models
(see Sect. 3.4.3.2) and ‘normal’ interval generation approach (see Sect. 5.6.6). The
best performance concerning the interval stability after disturbance shows linear
regression and auto arima the lows performance shows theta and arima (0.2.2) mod-
els. However, linear regression shows the largest interval and feting error.

Model Inputs Uncertainty Analysis of the Distant Domain

The main inputs to build the reference impact matrix are based on the literature
research and discussion with experts to check the relevancy of different categories,
such as trends, discontinuities, weak signals, and wild cards for the given study.
However, most of these categories where already discussed and published previ-
ously by a large group of experts [3, 4].

6.2.2.4 Model Output Uncertainty

Model Output Uncertainty in the Neighbouring Domain

Model output uncertainty of data driven models, which is also model input uncer-
tainty for process driven models, are analysed and presented using different tech-
niques for interval analysis (see Appendix D). The implementation of different
interval analysis methods can give different results. Different interval analysis
methods are presented in Fig. 6.4 and compared. Most stable interval generation
methods we denote ‘normal’ interval generation method. Intervals generated by
method based on Chebychev’s inequality generate very large non stable intervals
see Fig. 6.4.
The overall uncertainty of modelling the neighbouring domain includes not
only interval analysis but also uncertainty in the model input and model structure.
The process of uncertainty analysis of model output in the neighbouring domain is
used as model input uncertainty analysis of ‘process driven models’.

Model Output Uncertainty in the Targeted Domain

The model output of the targeted domain developed using process driven models
considers all uncertainty in different modelling steps and domains. Uncertainty
148 6 Implementation of Discussed Uncertainty Analysis Approaches …

Fig. 6.8  Consistency diagram of different scenarios

analysis for this Use Case—I is performed using the fuzzy scenario uncertainty
analysis approach and the output present in different scenarios.
The final scenarios using consistency analysis are presented in the Fig. 6.8.
Over 2000 sets of different key descriptor combinations are generated. However,
some of them are not consistent, and others have very low consistency values.
A priori and posterior probabilities of different descriptor evaluations using
CIM and fuzzy inference are presented in Fig. 6.9.
The results of analysis show that scenario number 2 from enumerated 200 sce-
narios has the highest consistency in the total set of scenarios.
Different scenarios have been developed based on the method discussed before.
The results show that scenario N 2 also has, in the fuzzy inference based scenario
generation, the highest frequency of appearance. However, such scenarios as N 1,
3 or 4, which have high consistency, had low values for appearance.
For the final selection of scenarios, fuzzy-cluster analysis is implemented (see
Sect. 5.6.10).
The initially generated number of raw scenarios was greater than 2000. The
first 200 most-consistent scenarios have been selected for cluster analysis. Low
consistent and non-consistent raw scenarios have not been considered in the
analysis.
6.2 An Example of Use Case I: Singapore 149

Fig. 6.9  A priori and

posterior probabilities of
different descriptors values

Each scenario has a certain value for several descriptors, which is also called
the dimension. In the Singapore Use Case, the scenarios have more than 10 dimen-
sions (key descriptor values). These dimensions have different scales and units,
e.g., population or energy costs. To normalise and bring different units or attrib-
utes into the same scale, normalisation has been performed using the following
formula:
ϑ − averagevalue
ϑ′ =
maxϑ − minϑ
ϑ is the real value
′
ϑ is the normalized value
The Fuzzy C mean (FCM) algorithm has been implemented (see Sect. 5.6.10).
150 6 Implementation of Discussed Uncertainty Analysis Approaches …

Fig. 6.10  The number of Objective Function

clusters versus the objective
120.00
function value
100.00
80.00
60.00
Objective
40.00
Function
20.00
0.00
2 3 4 5 6 7 8 9
Number of Cluster Centers

The analysis was initiated with 9 cluster centres. Figure 6.10 shows the decreas-
ing value of the objective function in relation to the amount of clusters. However, it
is not possible, on the basis of this trend, to select the final number of relevant clus-
ters because minimum optimality indicates the highest number of clusters.
The relation between the partition coefficient, the classification Entropy and the
number of clusters is presented in Fig. 6.11. To identify amount of clusters having
internally high homogeneity and externally high heterogeneous quality criteria, the
partition coefficient and classification entropy have been implemented.
For the analysis of the quality criteria, the coefficient and classification entropy
show a light elbow criterion for 3 clusters. This will be the orientation for the selection
of the number of clusters. Many studies recommend not using a large amount of sce-
narios, and instead limiting the number of scenarios within the range 3–5.
The scenarios that having the highest consistency value shows following mem-
bership values to different three clusters see Table 6.6.
The analysis shows that scenario N° 2 is also the most probable scenario. Based
on the cluster quality criteria (see Fig. 6.11), taking into account the analysis from
the previous sections. Scenarios having enumeration number N°2, N°176 and N°4
are the three scenarios considered for the final analysis. It can be interesting to
identify scenarios in opposite way, identifying first cluster centres and then defin-
ing scenarios. Doing so, it is possible to have high membership values than for
predefined scenarios.
The Singapore scenarios are visions that provide an opportunity to explore future
situations, to test implementation of actions and strategies and to support objective
and target reformulation. Different single sub models can be developed, and the out-
put uncertainty of these models can be evaluated according to these overall scenarios.
Each identified scenario has high consistency and stability and represents 3 dif-
ferent clusters.
Values of some descriptors in different scenarios are synthesised in Table 6.7.
Scenario Id. Nr. 2 Reference Scenario
This scenario has not only very high consistency but also highest frequency of
appearance compared to other scenarios. The developments of most of the key
descriptors have average value. Typical characteristics of this scenario are high com-
fort orientation, average electricity prices growth and average economic growth.
6.2 An Example of Use Case I: Singapore 151

Fig. 6.11  Trade off among the partition coefficient and classification entropy according to
amount of cluster centres

Table 6.6  Scenarios having high membership values to final cluster centres

Cluster id. Max. membership Id. number of scenario having highest
number value membership value
1 0.58514655 2
2 0.491060978 176
3 0.652417389 4

Table 6.7  Different scenarios for Singapore future development

Id_Nr. of scenarios 2 4 176
Consistency 42 40 34
Year 2020 2030 2020 2030 2020 2030
Outdoor temperature 29 29.5 29 29.5 29 29.5
Electr. price (¢/kWh) 41 55 37 50 34 45
Life style: high, low comport oriented High High High High Low Low
Solar radiation (kWh/m2/d) 4.5 4.5 4.5 4.5 4.5 4.3
HH income (S$) 14,200 18,000 14,200 18,000 10,800 14,000
HH size (Pers./HH) 3.52 3.52 3.47 3.47 3.47 3.47
Inhabitants 15<>64 74 77 73 74 73 74
Inhabitants <14 12.4 6.8 12.4 6.8 14 10.3
Inhabitants 65+ 14.5 20 14.5 20 13 15
GDP thousand (S$) 439 680.6 577.5 884.7 411.2 476.5
Ownership air-condition (%) 77 77 77 77 77 80
152 6 Implementation of Discussed Uncertainty Analysis Approaches …

Scenario Id. Nr. 4: High economic growth

Here, a significant increase of economic activities is expected in Singapore, com-
pared to other scenarios. For example, GDP is increasing, and electric price does
not increase strongly.
Scenario id. Nr. 176 Moderate economic growth
Here, economic and income growth is low. The share of children increase and
other descriptors do not show strong development.

Model Output Uncertainty in the Distant Domain

The output of the distant domain is presented in different categories, such as weak
signals or wild cards Table 6.2. Uncertainty of occupancy of these categories,
which might have impact on other models, is performed based on literature review.

6.3 An Example of Use Case II: Mexico City

The main study objective was the evaluation of various potential technologies for
energy production using waste in Mexico City. The model, which was developed
by the EIFER team with cooperation’s partners, includes the whole waste manage-
ment system and the potential energy system for energy production.
Contrary to Use Case I—Singapore, in this study the objective and initial solu-
tion were formulated a priori. Therefore, analysis of planning phase I according to
Sect. 2.2 in Chap. 2 was not required. Additionally, decision and context framing
uncertainty were not required.
This Use Case had an integrated, quantitative model that models different sec-
tors, such as waste, partial transportation and the energy system, in an integrated
manner. It was necessary to perform uncertainty analysis for this model. The
PRSUA approach was implemented for this model. The analysis steps and meth-
ods used were performed according to Sect. 5.5.3. Before uncertainty analysis
3-domain modelling concept was implemented (see Sect. 3.3).

6.3.1 Modelling Mexico City’s Waste-to-Energy System

Modelling the distant and neighbouring domain and the methods implemented are
the same as in Use Case I—Singapore. However, modelling of the targeted domain
is different. The demonstration of the modelling of the targeted domain is con-
ducted in the following sections.
6.3 An Example of Use Case II: Mexico City 153

6.3.1.1 The Waste Management System in Mexico City

Mexico City, also called Distrito Federal (D.F.), is one of the 32 federal enti-
ties of the Mexican State. It is the capital of Mexico and, together with the Zona
Metropolitina del Valle de Mexico City (ZMVM), is one of the most populated
domains in the world. The ZMVM consists of the 16 delegations, or boroughs,
of Mexico City, 59 municipal domains of Mexico State and one of the states of
Hidalgo. Approximately 20 million people live in the ZMVM, with 8.8 million liv-
ing in the D.F.
Mexico City generates, by itself, an amount of ca. 12,500 tons of solid waste
per day, of which 60 % correspond to inorganic and 40 % to organic waste. The
distribution of waste generation per sector is shown in Fig. 6.12. Almost 50 % of
waste is generated in dwellings, while the commercial sector accounts for 29 %
and the service sector for 15 % [8]. Waste recycling is only of minor importance
in Mexico City: only between 3 and 10 % of the total waste generated is recy-
cled in one of the three selection plants; the rest is disposed of in the landfill site
“Bordo Poniente”. Although “Bordo Poniente” is currently the only landfill site,
it was planned to close it already by the end of 2013. It is expected that the rate
of waste production will increase in Mexico City by 130 tons per year; therefore,
environmental management will become an important issue for the government
[9]. Several projects and action plans have been suggested and are in-progress
for implementation, such as a new landfill site or so-called “Centro Integral de
Reciclaje y Energia (CIRE)” on waste transfer to a landfill site in “Mineral de

Fig. 6.12  The waste management system in Mexico city

 154

Total Waste
Production <perc c3-t3>

Waste in Sorting and MV Reuse (m1)

<RATIO OF WASTE <RATIO OF WASTE Recycling Center (t3)
<GROWTH RATE FROM EMPLOYEE PER
FROM INHABITANT
WASTE PER YEAR> <perc c3-t5>
PER YEAR>
INHABITANT> Waste in DtD
Selective Collection Waste in Compost MV Recycling
Waste Per Center (t5) (m2)
(c3)
increase in waste Inhabitant
per inhabitant <perc c3-t4>
6

<perc 0-c3> Waste in

<Time> Anaerobic MV Compost
Waste Production Digester (t4) (m3)
<GROWTH RATE <NUMBER OF
Per Year (w)
INHABITANTS> INHABITANTS> <perc c4-t6>
Waste in Mechanical EV RDF (e1)
Inhabitants Biological Tretment
<perc 0-c4> Center (t6)
increase of
inhabitants <perc c4-t8>
Waste in DtD
Waste in EV Heat (e2)
<NUMBER OF Non-Selective
Collection (c4) Incinerator (t8)
GROWTH RATE TERTIARY
TERTIARY EMPLOYEES>
<perc c4-t10>
EMPLOYEES
Waste in EV Biogas (e3)
Tertiary Landfill (t10)
increase of Employees
tertiary
employees
Liquid
Waste in Home Residues (r1)
<NUMBER OF <THERMAL Composting (t1)
INDUSTRIAL TECHNOLOGIES
GROWTH RATE EMPLOYEES> <perc 0-t1>
CENTER Y/N>
INDUSTRIAL Waste in
EMPLOYEES Uncontrolled Solid Sewage Sludge
Industrial Disposal (t2) Residues (r2) Land Spreading (t7)
Employees
increase of <perc 0-t2>
<THERMAL
industrial TECHNOLOGIES <perc r1-t7>
employees CENTER Y/N> Waste
Non-Valorised

Fig. 6.13  System dynamics model of Mexico city’s “Waste-to-energy” model [11]

Implementation of Discussed Uncertainty Analysis Approaches …
6.3 An Example of Use Case II: Mexico City 155

Reforma”, in the state of Hidalgo. Within the special program for climate change,
two overall objectives were proposed: (1) achieve an integral management of
waste and (2) avoid non-controlled emissions in final disposal sites by means of
controlled combustion and energy generation through waste utilisation [10].

6.3.1.2 Modelling the Targeted Domain and Identification

of Key Descriptors

According to the study and project needs, the targeted domain only focuses on
the energy, waste and partially transportation system necessary to process waste
from the time of collection until and its energetic use. System dynamics modelling
approach, which describes the system as processes, was implemented (Fig. 6.13).
Because of confidential reasons the system dynamic model won’t be discussed
here detailed. The main outputs, or key performance indicators, of the waste-to-
energy model have been identified at the beginning. They are the total produced
waste, total costs and total GHG emissions. For identification of the main model
driver or descriptors, a one-at-a-time sensitivity analysis approach has been used
(see Sect. 5.6.11). Each model input has been changed separately keeping the rest
of the model inputs unaffected intentionally (Fig. 6.14).
From hundreds model inputs, 15 key model drivers have been identified as hav-
ing the highest impact on the model outputs (Table 6.8).

6.3.1.3 Modelling Neighbouring Domain

The main drivers of the waste-to-energy system have been modelled using data
driven modelling approaches. Different methods and models, such as ARIMA (0.2.2)
or neural nets, are discussed in Sect. 3.4.3. Different data analysis techniques exist
for historical data analysis. The focus of this book is not the descriptive but induc-
tive analysis (Fig. 6.15; Table 6.9). However, some results of descriptive analysis like
decomposition of time series, identification of outliners are presented in Appendix A.
The analysis (see Table 6.3) shows that, for large amount of memory data or
available observation (more than 12 historical periods), for 24 data sets from dif-
ferent domains, including demography and economy, the best MASE value has
ARIMA (0.2.2), automatic ARIMA. The best MdAPE values show automatic
ARIMA and ARIMA (0.2.2). More results of the analysis are in Appendix C.
The final evaluation of data driven models based on both use cases are provided
in Chap. 7.

6.3.1.4 Modelling Distant Domain

A reference impact matrix has been implemented for modelling the distant domain.
The same categories implemented in Use Case I—Singapore, such as wild card or
156 6 Implementation of Discussed Uncertainty Analysis Approaches …

Fig. 6.14  Tornado diagram of Mexico city’s waste-to-energy model descriptors

Table 6.8  List of the 15 key descriptors of Mexico city waste to energy model
Related model output Key model drivers or descriptors
Total waste Ratio of waste from inhabitant per year_Residential_glass
Growth rate of inhabitants
Growth rate waste per inhabitants
Total GHG Fuel consumption per truck diesel
Global emission per fuel diesel
Ratio of waste from inhabitant per year_Residential_SC
Global emission per fuel_LPG
Emissions Global elect. grid
Total cost Ratio of waste from inhabitant per
year_Residential_bulky
Specific operating costs sorting centre
Ratio of waste from inhabitant per year_Residential_SC
Specific operating costs landfill
Specific investment costs sorting centre
Specific investment costs landfill
Specific costs manpower and equipment

weak signals, have been implemented. The expected global trends of wild cards
and weak singles were similar to the Singapore case. However, the implementation
of RIM was not completely finalised due to the lack of local information in Mexico
City and project constrains. The global trends, discontinuities, wild cards and weak
signals presented in Table 6.2 are also usable for Mexico City.
6.3 An Example of Use Case II: Mexico City 157

Fig. 6.15  Population growth in Mexico, multi model simulation, composite extrapolation (Data,
historical data; Theta, ARIMA (0,2,2), ARIMA (4,1,2) is generated via auto.arima; NNAR(1),
neural networks; RandWalkFwD, random walk with drift; LinReg, Linear Regression, RobTrend,
Robust trend [see more details in Sect. 3.4.3])

6.3.2 Uncertainty Analysis

6.3.2.1 Context and Framing Uncertainty Analysis

As was mentioned previously, the context and the frame, as well as the objective
and solutions or actions using waste for energy production, was predefined for
this study a priory. Therefore, the methodology discussed in Sect. 5.6.4 was not
implemented completely. However, the implementation of conformity checklist [6]
helps to updated the model context and boundary conditions.

6.3.2.2 Model Structure Uncertainty Analysis

Introduction

The discussion of model structure uncertainty is concentrated on the process

driven model structure uncertainty employed for the targeted domain. Model
structures for uncertainty of the neighbouring and distant domain are similar to
those of Use Case I—Singapore, and therefore it won’t be discussed fully here.
158 6 Implementation of Discussed Uncertainty Analysis Approaches …

Model Structure Uncertainty Analysis in Targeted Domain

Dimensional consistency and extreme conditions tests have been implemented

to test the model structure of the waste-to-energy system dynamics model (see
Sect. 5.6.8).
The first test implemented is the dimensional consistency test to check the con-
sistency of dimensions. This makes possible to review the consistency of equa-
tions. Also the syntax has been checked, scaling issues and warnings arising from
Lookups that have gone out of range and from variables that have gone beyond
their specified low/high range see Fig. 6.16.
The next test, which can be observed as a part of the sensitivity analysis, is the
extreme conditions test. Giving different extreme values to model inputs, the out-
put behaviour of the model has to be verified.

Fig. 6.16  Syntax, and model components specification tests of Mexico city’s waste-to-energy
model

Table 6.9  Average performance of data driven models and methods across all data sets for some
data of Mexico
Methods, Theta ARIMA auto.ARIMA Nnet LinReg Random RobTrend
models (0.2.2) walk with
drift
MASE 0.975 0.492 0.514 1.129 1.9 0.78 0.76
MdAPE 0.032 0.019 0.018 0.022 0.046 0.024 0.023
Theta model, arima (0.2.2) model, arima (1.2.2) is generated via auto.arima; nnetar, neu-
ral networks; RandWalkFwD, random walk with drift or ‘trend’; LinReg, Linear Regression,
RobTrend, Robust trend (see more details in Sect. 3.4.3)
6.3 An Example of Use Case II: Mexico City 159

An example shows that if Municipal Solid Waste Generated per capital is zero,
then the total produced municipal waste must also be zero.
Other tests like the policy implication test that examine how robust policy
implications are when changes are made in boundaries or parameters, are not
implemented at this stage.

6.3.2.3 Model Inputs Uncertainty Analysis

Model Inputs Uncertainty Analysis of the Neighbouring Domain

Model inputs uncertainty analysis in neighbouring domain is performed similar

to Use Case I—Singapore see section “Model Inputs Uncertainty Analysis of the
Neighbouring Domain”.

Model Input Uncertainty Analysis in the Targeted Domain

Uncertainty of Individual Model Driving Forces

The analysis of model inputs concentrates on 15 selected key descriptors iden-
tified in the previous section. For probabilistic sampling-based approaches, it is
necessary to define the probabilistic distribution function (PDF) of different input
parameters. The work on the Mexico City study shows that the availability of data
is very low. Only some demography data, such as population, and economic data,
such as employment, were available. The rest of the inputs have been defined
using expert judgment and reference values from European cities. Therefore, the
definition of the PDF was extremely difficult, and it is based only on expert judg-
ment using common PDF functions, such as uniform, normal or log-normal dis-
tributions. For example, the ratio waste from inhabitant per year is implemented
by making analogies to European cities because of the missing Mexican data.
The Global CO2 emissions are selected as an average of South American coun-
tries. Uncertainty of other model variables and descriptors are not provided here
because of study confidential issues (Fig. 6.17; Table 6.10).

Uncertainty Because of Interdependency Among Different Model Inputs

and Linguistic Uncertainty
To analyse interdependency among different model inputs SyntheSim simulation fea-
ture of Vensim software is used. This allows exploring model response to wide range
of model inputs assumptions combination. Possible interdependency is then evalu-
ated in graphical way interactively. Because of large number of inputs the focus was
to explore the interdependency among key descriptors. More formal approach dis-
cussed in Sect. 5.6.9 such as copula is not implemented in this Use Case yet.
160 6 Implementation of Discussed Uncertainty Analysis Approaches …

Fig. 6.17  Description of uncertainty of some model inputs

Table 6.10  Description of uncertainty of some model inputs

Model input Description Values
Ratio waste from inhabitant per year Min value 250
(kg/cap/a) Max value 800
Mean 480
Standard deviation 113
PDF Normal distribution
Global emissions for el. production Min value 102
(CO2 t/(kWh/a)) Max value 177
PDF Uniform

Model Input Uncertainty Analysis in the Distant Domain

Model inputs uncertainty analysis in distnat domain is performed similar to Use Case
I—Singapore see section “Model Inputs Uncertainty Analysis of the Distant Domain”.

6.3.2.4 Model Output Uncertainty Analysis

Model Output Uncertainty Analysis in Neighbouring Domain

Model outputs uncertainty analysis in neighbouring domain is performed simi-

lar to Use Case I—Singapore see section “Model Output Uncertainty in the
Neighbouring Domain”.
6.3 An Example of Use Case II: Mexico City 161

r03
50% 75% 95% 100%
Total Waste Production
200 M

150 M

100 M

50 M

0
0 5 10 15 20
Time (Year)

Fig. 6.18  Model output uncertainty for the total waste generated

Model Output Uncertainty of Process Driven Models

of the Targeted Domain

Model output uncertainty is generated using Monte Carlo simulation. The results
of model outputs are presented using intervals and the distribution of values with
quintiles (Fig. 6.18).
The results show a decreasing growth of the central value of the total waste pro-
duced in Mexico City. However, the uncertainty analysis shows that this value has
high uncertainty and e.g., in 20 years the waste production could ¾ time more than
central value.

Model Output Uncertainty Analysis in Distant Domain

Model outputs uncertainty analysis in neighbouring domain is performed similar

to Use Case I—Singapore see section “Model Output Uncertainty in the Distant
Domain”.

References

1. Energy Supply Security Planning in the ASEAN (ESSPA), Promotion on Energy Efficiency
and Conservation (PROMEEC) (2009). The 2nd ASEAN Energy Demand Outlook
2. Brief diagnosis of the energy and environment context of Singapore (2010). EIFER-
Technical report, Karlsruhe
162 6 Implementation of Discussed Uncertainty Analysis Approaches …

3. Saritas O, Smith JE (2011) The big picture—trends, drivers, wild cards, discontinuities and
weak signals. Futures 43(3):292–312. doi:10.1016/j.futures.2010.11.007
4. Saritas O, Nugroho Y (2012) Mapping issues and envisaging futures: an evo-
lutionary scenario approach. Technol Forecast Soc Chang 79(3):509–529.
doi:10.1016/j.techfore.2011.09.005
5. Wong T-C, Yuen BKP, Goldblum C (2008) Spatial planning for a sustainable Singapore.
Springer, Singapore
6. Mirakyan A, De Guio R (2014) A methodology in innovative support of the integrated
energy planning preparation and orientation phase. Energy 78:916–927. doi:10.1016/j.
energy.2014.10.089
7. Sinus Software and consulting. http://www.sinus-online.com/szeno-plan-top.html
8. Inventario de Residuos Solidos del Distrito Federal (2010). (3)
9. de Buen Rodríguez O (2008) Acción climática de la ciudad de México 2008–2012
10. Programa Especial de Cambio Climático 2009–2012 (2009)
11. Susanne Schmidt et al (2012) Mexico “Waste to energy” model. Technical report. EIFER,
Karlsruhe
Chapter 7
Evaluation and Discussion

The objectives of this chapter are

• Provide feedback on the implementation of different modelling methods in
­different domains according to quality factors
• Discuss the evaluation of PRSUA and FSUA according to quality factors
• Present trades-off among PRSUA, FSUA and a commonly used deterministic
approach according to quality factors

7.1 Evaluation and Discussion of the 3-Domain Modelling

Concept and Different Modelling Approaches

7.1.1 General

The 3-domain modelling approach, which is developed based on the 3-domain

Metasystem framework, uses different modelling approaches for different model-
ling domains. The evaluation and discussion of different implemented modelling
approaches are discussed below according to quality factors identified in Sects. 2.6
and 5.3. However, only quality factors such as technical, organisational and satis-
faction by planning participants are considered for the evaluation. Quality factors,
such as collaboration support or knowledge and discovery in the sense of avoiding
mental inertia, are not discussed in this book. Nevertheless, relevant references for
these quality aspects and assessment of the methods are referred.

© Springer International Publishing Switzerland 2015 163

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_7
164 7 Evaluation and Discussion

7.1.2 Modelling Approaches for Targeted Domain

7.1.2.1 Technical Quality

The main technical quality of modelling approaches in targeted domain is the

representation of real system taking into account the special planning condi-
tions. The implementation of system dynamics approach for modelling waste to
energy system in Use Case-II provided highly satisfactory results. System dynam-
ics makes possible to incorporate multiple dynamic interactions feedbacks of dif-
ferent system elements. However, there remain some limitations using system
dynamics in context of city or territory energy planning, which are discussed in
Sect. 3.3.3.1.
MICMAC approach was implemented in Use Case-I. Despite low data avail-
ability and heterogeneous requirements in this case, MICMAC gives the opportu-
nity to model the city as a whole on qualitative level. It makes possible to see the
dynamics of main city descriptors and incorporate non prices qualitative informa-
tion. MICMAC allows also representing the influence rank of different descriptors
and select key drivers of city.

7.1.2.2 Organisational Capability

Both MICMAC and system dynamics are flexible approaches and do not require
high amount of time and resources. Where as MICMAC approach is qualitative,
system dynamics required primarily quantitative data. For MICMAC there is a
need for expertise in order to represents interaction among different descriptors,
for system dynamics there is also a need of knowledge to represent real physical
and/or economic processes in form of equations, relations etc.

7.1.2.3 Satisfaction by Planning Participants

Using MICMAC approach was very suitable because of its transparent easy under-
standable logic that takes into account special planning conditions with low quan-
titative data availability. Despite of the scarcity of data, it provides some results
required by planning participants.
As system dynamics is a commonly used approach, planning participants
were familiar with the method and no need to additional learning was required.
Particularly, system dynamics makes it possible to have a quantitative aggregated
view of the long range system’s behaviour, which was required in Use Case-II.
7.1 Evaluation and Discussion of the 3-Domain Modelling … 165

7.1.3 Modelling Approaches for Neighbouring Domain

7.1.3.1 Technical Quality

Data driven modelling approaches were selected on the base of the literature review.
After implementing them in Use Cases, the methods show different performances
according error measure MASE and MdAPE discussed in Sect. 5.6.9. The compari-
son of seven data driven modelling approaches for inductive analysis, for extrapo-
lation, performed on 53 demographic, economic, climate etc. data sets, shows that
automatic ARIMA and ARIMA (0, 2, 2) have the best performance according to
MASE and MdAPE. Theta model and linear regression shows the lowest perfor-
mance. The average performance according to MASE and MdAPE shows neural
networks, random walk with ‘drift’ and Robust trend. However, the purpose of mod-
elling the neighbouring domain is not only seeking “the best” modelling method
with the least error, but having a set of different methods that can be used for com-
posite inductive analysis, triangulation and model structure analysis using different
methods from different paradigm. The best set of methods, according to the need
identified in Sect. 2.6 for long range planning, can be the composed of auto ARIMA,
ARIMA (0, 2, 2), neural networks and Random walk with drift (or we denote with
‘trend’) as reference because of its simplicity and above-average performance across
all data sets. In addition to trends identification and extrapolation, Auto ARIMA and
neural networks can also discover cyclical effects for annual based data sets.

7.1.3.2 Organisational Capability

Data driven modelling approaches are quantitative and require different data
amount for appropriate analysis. As discussed in Sect. 3.4.3 ARIMA approaches
and particularly neural networks method require sufficient amount of data to be
effective. In scare data conditions such as diffusion of new technologies logistic
S-shaped curve method was very useful. The simplicity of modelling approaches
as discussed is important. ARIMA approaches and neural networks required cer-
tain analytical expertise. Other methods like random walk with drift or linear
regression are easier to understand or use, but they have low technical perfor-
mance discussed above. Most of the tested data driven modelling approaches are
available in statistical software’s like R.

7.1.3.3 Satisfaction by Planning Participants

The selected data driven approaches were either predefined i.e. there was no need to
adapt the model to data, or the model was automatically adapted to give data set like
auto ARIMA with less subjective judgment. Automatic features made the approach
reproducible, very flexible, easy to implement for different data sets and provides
166 7 Evaluation and Discussion

satisfactory results. However, the implementation shows that the planner needs cer-
tain analytical background or ability to learn new approaches like neural networks,
which is not commonly included in regular engineering studies at the universities.

7.1.4 Modelling Approaches for Distant Domain

7.1.4.1 Technical Quality

Usually, distant domain is not considered in infrastructure planning explicitly. The

discussed reference impact matrix (RIM) can model, present and organise infor-
mation in distant domain. It gives opportunity to analyse the models of city or ter-
ritory jointly with different global or national trends, discontinuities etc. instead of
separately.

7.1.4.2 Organisational Capability

For development of RIM on the global level there was no additional information
required, since some information about trend, weak signals etc. was published by
expert groups before. Expertise was required for information on national and local
level which requires some time and knowledge. RIM was flexible to implement in
both Use Cases.

7.1.4.3 Satisfaction by Planning Participants

Implementation of RIM was considered as satisfactory by the planner. Having

additional information outside to study boundaries increase the satisfaction of
planning participants and readability of models for decision makers.

7.2 Evaluation and Discussion of Uncertainty Analysis

Approaches

7.2.1 General

Quality factors for the uncertainty analysis methods were discussed in Sect. 5.3
for evaluation of the discussed uncertainty analysis approaches these quality fac-
tors are utilised. However, discussion from the literature is included for an overall
evaluation of uncertainty analysis methods. For some quality factors, such as sat-
isfaction with methods, the evaluation was performed using a qualitative descrip-
tion on the basis of the experience implementing them in the given Use Cases and
literature research.
7.2 Evaluation and Discussion of Uncertainty Analysis Approaches 167

7.2.2 Evaluation of FSUA Multi Method Approach

and Discussion

7.2.2.1 Technical Quality of FSUA

Completeness of the Incorporation of Uncertainties

After implementing FSUA in Use Case I—Singapore, the key points can be char-
acterised as follows:
• FSUA can address almost all types of uncertainty except total ignorance.
• It considers not only risk situation but also uncertain situation.
• Linguistic uncertainty is incorporated using Fuzzy linguistic terms, which
increase also the incorporation of non-precise information.

Incorporating Qualitative and Quantitative Information

The FSUA approach can deal with both qualitative and quantitative information.
Qualitative information can be transformed using Fuzzy linguistic terms and infer-
ence approaches to obtain crisp quantitative values for implementation in the next
steps of analysis.
Automatism
The development of scenarios does not have any automatism. When the situation
changes or new information is available, scenarios have to be redeveloped or refor-
mulated again. In Use Case I—Singapore, the scenarios were developed at the end
of the study, once all possible information was included and no changes or adapta-
tion of the scenarios was required. In future research, the implementation of a par-
tial automated approach suggested by [1] could speed up the scenario generation
process.
The intervals with data driven models are high automated methods. In case
when new data or information is available the methods can automatically perform
multi method simulations with interval providing composite extrapolation which
allows making triangulation of results.
Mixing Synergy
FSUA is a multi-method approach. It uses probabilistic and fuzzy set theory in
combination. The approach is based on explorative, inductive scenario analysis.
Different methods are used in different planning and modelling steps (see Sect.
5.5.2). The design of FSUA, similar to the PRSUA approach, is performed using
different mixing rationales. Rationales, such as triangulation, are utilised for ana-
lysing model structure uncertainty, or extension for generating information for
other methods.
168 7 Evaluation and Discussion

7.2.2.2 Organisational Capability

Required Time and Resources

For implementation of FSUA, the time and resources needed are low or middle
for data collection (see Sect. 5.3). Particularly for scenario generation, which
requires expert elicitation to define scenarios and conduct consistency and cross-
impact analysis, the time and resource requirements are very high. In Use Case
I—Singapore, three different meetings were organised for expert elicitation, each
for around 2 h excluding preparation and the involvement of experts.
The level of information required can be ‘Int. 01’ to ‘Int. 02’, according to the
classification discussed in Sect. 5.3. Implementation of the FSUA approach in the
Singapore case shows that using available quantitative and qualitative information
makes it possible to develop sufficient scenarios within the given time frame of
7 months.
Required Expertise and Skills
The scenario based approach is the most popular approach in infrastructure plan-
ning. There is no specific skill or knowledge required to develop the scenario.
However, Fuzzy inference and Fuzzy set theory are not commonly used and
require some knowledge and experience. Fuzzy set theory is very intuitive and
close to the human understanding and expression of knowledge.
Flexibility
FSUA as implemented in Use Case I—Singapore can also be implemented for the
Use Case II—Mexico because it can be used independently from the availability
of quantitative models of the targeted domain.
Legitimacy
There is no information available at this stage of the research about regulation or
directives for implementation of the certainty approach for uncertainty analysis in
cities and territories infrastructure development planning. Therefore, this aspect
could not be evaluated.

7.2.2.3 Satisfaction by Planning Participants

Satisfaction Working with Method

Based on FSUA, scenarios have been developed despite low data availability. Experts
working with the method found it sensible and straightforward. Scenario develop-
ment is not necessarily linked with probability information, which makes it possible
to collect and develop initial scenarios results during the project. FSUA makes work-
ing with the method very intuitive and easy for expressing the knowledge and incor-
porating the information using linguistic terms and non-precise information.
7.2 Evaluation and Discussion of Uncertainty Analysis Approaches 169

Satisfaction of Decision Makers with Regard to the Method

The scenario approach is one of the frequently used approaches. The first results
of the FSUA approach implementation in Use Case I—Singapore were com-
municated to project participants. It received high acceptance mainly because of
transparency of method and scenarios development in given time frame when data
availability was low. However, the experience in the Singapore case shows that
understanding and use of scenario terminology can differ with planning partici-
pants. This can cause misunderstandings of the method. This situation is not unique
because, as stated by [2], the scenario approach is a frequently used and misused
technique; therefore, it can be helpful to organise workshops that will explain ter-
minologies and methods to all planning participants and decision makers.

7.2.3 Evaluation of PRSUA Multi Method Approach

and Discussion

7.2.3.1 Technical Quality of PRSUA

Completeness of the Incorporation of Uncertainties

PRSUA incorporates different typologies of uncertainty. The key points can be
characterised as follows:
• The PRSUA approach can address many type of uncertainty.
• It focuses on probability theory and requires a sufficient amount of data availability.
• Levels ‘uncertainty’ or ‘ignorance’ cannot be addressed using PRSUA alone.
• Linguistic uncertainty cannot be addressed using PRSUA without additional
methods outside probability theory.
Incorporating Qualitative and Quantitative Information
The PRSUA approach is mainly quantitative. Qualitative information must be
transformed using additional approaches, such as maximum entropy principle [3],
to identify appropriate probability density functions of certain input parameters or
driving forces. This approach is not implemented.
Automatism
PRSUA has high automatism as it can create random samples automatically when
new information is available and models have to be updated. However, there is a
need to incorporate interdependency among model inputs by using conditional dis-
tribution or another approaches, such as Copula. This will require expert knowledge.
Mixing Synergy
PRSUA is a multi-method approach which employs diverse methods in differ-
ent planning and modelling steps. Its individual methods are implemented using
different mixing rationales, such as triangulation for increasing model structure
170 7 Evaluation and Discussion

uncertainty or development to generate information, such as using one-at-a-time

sensitivity analysis to identify key descriptors for future analysis. The use of dif-
ferent rationales like triangulation or extension increase advantages using PRSUA.

7.2.3.2 Organisational Capability

Required Time and Resources

For complete implementation of PRSUA, the time and resources needed are high
for data collection. The level of information required for PRSUA is at least ‘Int. 04’
or ‘Int. 05’: Probability distribution of data and relevant measures”, according to
the classification discussed in Sect. 5.3.2. This is the heights level of information
needed for full functional analysis. The Mexico City use case shows that the waste-
to-energy model required over 100 s different input parameters and drivers. After
filtering, there remained still 15 key descriptors, which is pretty high. An empirical
probability density function for only 8 model inputs out of the 15 parameters, such
as demography and employment was found.
Required Expertise and Skills
The implementation in the Use Case—II shows that some explanation and inter-
pretation of the terminologies and methods was required for most of the planning
participants having non-mathematical background. Therefore, it is necessary to
inform and guide planning participants by using PRSUA.
Flexibility
One of the reasons why the PRSUA approach with random sampling was only imple-
mented in the Use Case II-Mexico was that it requires a quantitative model for the
targeted domain. In the Singapore case, there was no integrated quantitative model
available at the beginning, which made the use of PRSUA impossible. Indeed, when
there is no quantitative model available or developed, the separate use of PRSUA for
the different partial quantitative models is technically possible but when links among
them are not known, the overall uncertainty assessment cannot be performed.
Legitimacy
To date and to our knowledge, there is no information available regarding regu-
lation or directives for implementation of the certainty approach for uncertainty
analysis in IEP in cities and territories.

7.2.3.3 Satisfaction by Planning Participants

Satisfaction Working with Method

Because of increasing computing power and number of software having random
sampling functionality, random sampling has become increasingly popular. In par-
ticular, the automatic random sampling procedure makes this approach popular.
Table 7.1  Comparative assessment of proposed and commonly used uncertainty analysis approaches
7.2

Requirements Quality factors Commonly implemented PRSAU FSUA according degree 4b

degree 3 “best estimate” according
degree 4a
Technical quality factor Extend of incorporation Deterministic view, no Probabilistic view, most All defined uncertainty
of uncertainty typologies uncertainty analysis is done uncertainty typologies are typologies are incorporate
explicitly incorporated except levels: except total ignorance
uncertainty, ignorance and
linguistic uncertainty
Incorporating Quantitative Quantitative Qualitative and quantitative
qualitative and
quantitative information
Automatism High High using automatic Low, there is a need to check
random sampling technique the scenario consistency using
quantitative information or
experts judgment if situation
change
Mixing synergy Not considered explicitly Considered Considered
Organisational capability Min. required information Middle: int.03 High: int 04 or int 05 Low: mix of int.01 and 02
Required expertise Low High−middle High−middle
and skills
Evaluation and Discussion of Uncertainty Analysis Approaches

Flexibility High Low High

Legitimacy Not defined Not defined Not defined
Satisfaction by planning Satisfaction working Not know Middle Middle
participants with method
Satisfaction by DM Not know Not know High
with method
171
172 7 Evaluation and Discussion

However, In order to perform the automatic sampling procedure, the interdepend-

ency among random parameters still remains to be defined using the appropriate
approaches, which require additional expertise and time. Available amount of data
was too low to provide satisfactory results in Use case-II using PRSUA.
Satisfaction by Decision Maker with Method
PRSUA results were not communicated with the local decision maker in Use Case
II—Mexico City. However, the main concern of other planning participants was
that the approach could only be implemented and the results could only be com-
mented in probabilistic terms, when sufficient data or expertise are available. This
was not possible in Use Case II.

7.2.4 Comparative Assessment of Proposed Approaches

The review of different studies shows that, in context of integrated energy plan-
ning in cities and territories Sect. 4.3 and the analytical sophistication Degree 3
“Best estimate” method 5.2 is mostly used yet. However, in Sect. 4.1, the conse-
quence of using the deterministic “Best estimate” approach was shown, along with
reasons to be explicit for uncertainty analysis. Based on the discussion and analy-
sis of the implementation of FSUA and PRSUA in the previous sections, a qualita-
tive comparison of the discussed multi method approaches for uncertainty analysis
is presented in Table 7.1.
The conclusion of the comparative assessment is summarised in last Chap. 8.

References

1. Dönitz E, Möhrle PDMG (2009) Effizientere Szenariotechnik durch teilautomatische

Generierung von Konsistenzmatrizen. Gabler Verlag, Wiesbaden
2. Durance P, Godet M (2010) Scenario building: uses and abuses. Technol Forecast Soc Change
77(9):1488–1492. doi:10.1016/j.techfore.2010.06.007
3. Buckley JJ (1985) Entropy principles in decision making under risk. Risk Anal 5(4):303–313.
doi:10.1111/j.1539-6924.1985.tb00186.x
Chapter 8
Overall Conclusion and Future Research

The objectives of this chapter are

• Summarise and present the main outcomes
• Present detailed concluding remarks for the different chapters
• Present future work

8.1 Overall Synthesis and Conclusions

By addressing several open questions, the book presents the following main
­subjects. First, a 3-domain modelling concept is presented using a 3-domain meta-
system framework. Second, two novel multi-method approaches for uncertainty
analysis in the context of energy planning have been suggested. The implementa-
tion of these approaches have been demonstrated in Use Cases in Singapore and
Mexico cities and evaluated using quality factors for method evaluation.
Detailed findings and contributions are synthesized according to chapters of the
book below.

8.2 Synthesis and Conclusions of Chaps. 1 and 2

The extensive reviews in Chaps. 1 and 2 show that the following are general trends
in the context of integrated energy planning in cities or territories (IEPCT):
• Within cities and territories, a growing community awareness of environmental
issues;
• Growing interest in the use of distributed generation technologies based on
renewable resources and small-scale technologies such as cogeneration systems;

© Springer International Publishing Switzerland 2015 173

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8_8
174 8 Overall Conclusion and Future Research

• An increasing number of decision makers with different interests and prefer-

ences participating in the planning process; and
• The development of a cross-sector analysis among several sectors, such as
industry, households and transportation.
Because of these trends, the dynamics and complexity of energy infrastructure
planning tasks at the sub-national level have increased.
The review also shows the following:
• The planning and modelling process has some common procedures and can be
divided into different phases and sub-steps. However, this process is not sequen-
tial and can have iterations and omitted steps.
• Studies show that IEPCT is performed in uncertainty conditions; which means
that no probability information about the state of the environment is available or
probability can be only roughly defined.
• IEPCT is not flexible planning as defined in Sect. 1.3.1. Rather, it is rigorous
planning for the next 10–30 years, which does not lend itself to a quick modifi-
cation or response.
• The energy system in cities or territories is more of a socio-technical system,
which is more of a complex infrastructure, than a technical system alone.
• Multiple criteria, low data availability and limited budget conditions must be con-
sidered when planning and selecting a set of alternatives for cities and territories.
• Despite the importance and increasing popularity of incorporating uncertainty
analysis in other areas, very few city or territory studies have been performed that
explicitly report on analysing uncertainties in energy planning and modelling.
• Most model-based energy planning studies are deterministic. The analysis is
limited on Degree 3 of analytical sophistication, focusing on the central value
of model states. The consequence of using only a central or average value for
model parameters and variables can lead to large planning mistakes.
• Parametric uncertainty has mainly been addressed in only a few studies.
• Some studies state the need for integrated, holistic uncertainty analysis in the mod-
elling and planning process and not at the end when the plan or model is ready.

8.3 Synthesis and Conclusions of Chap. 3

The 3-domain Metasystem framework discussed in Chap. 3 allows identification

of an appropriate modelling approach considering not only the complexity of the
planning system consideration but also multiple study perspective and planning
constraints such as limited expertise or data. This allows a coherent, comprehen-
sive and pragmatic coverage of the planning problem. Based on the 3-domain
Metasystem framework, the 3-domain modelling concept is discussed. The mod-
elling methods show different performance according to quality factors. For dif-
ferent domains and Use Cases, different modelling methods are identified on the
basis of literature discussions and experiences using different modelling methods
in two different Use Cases.
8.3 Synthesis and conclusions of … 175

Suggested planning and modelling steps and a new three-domain modelling

concept make it possible to present different modelling and uncertainty analysis
approaches according to planning and modelling objectives, available expertise,
and data or resource availability, which differ in different planning and modelling
steps and domains.

8.4 Synthesis and Conclusion of Chap. 4

Based on integrated energy planning and modelling studies’ review, the need for
consideration of uncertainty is highlighted. Using an identified planning and model-
ling framework, a conceptual basis of uncertainty is presented showing the alloca-
tion of different types of uncertainty according to each planning and modelling step.
A review of practices in energy planning and modelling shows the gap between
need and practices and implemented methods to address some uncertainty types.
This conceptual basis clarifies the diverse typology of uncertainty in the plan-
ning and modelling process. This uncertainty concept can be reused in other
studies dealing with uncertainty, which is problematic not only in the context of
IEPCT but also in other model based infrastructure planning such as transportation
or water resource management.

8.5 Synthesis and Conclusions of Chaps. 5, 6 and 7

The development, implementation and evaluation of multi-method approaches for

addressing different types of uncertainties was the discussion focus in Chaps. 5, 6
and 7. Uncertainty analysis can have different purposes and degrees of analytical
sophistication. Two different degrees of analytical sophistication for uncertainty
analysis have been discussed according to IEPCT needs and compared to exist-
ing praxis. The highest analytical sophistication degree of uncertainty analysis in
the context of IEPCT, which requires more data and resources, is disputable and
requires further research.
New quality factors for the evaluation of uncertainty analysis methods have been
presented in Sect. 5.3 based on generic identified quality factors from Sect. 2.6.
Existing methods for uncertainty analysis have been reviewed and evaluated
using these quality factors. The evaluation shows that
• There are different methods for coping with different types of uncertainty.
However, none of them can address all types of uncertainty.
• There are comprehensive methodologies, such as NUSAP or PRIME, which can
be used as a framework for using different methods. However, these method-
ologies are focused primarily on environmental or process driven models and
not on data driven models, and the combination of different individual methods,
depending on Use Case condition, domains and planning or modelling proce-
dure, is not yet specified.
176 8 Overall Conclusion and Future Research

Based on this review, specific IEPCT conditions, identified analytical sophistica-

tion degrees, modelling or planning steps and domains and experiences of mixed
method research multi-method approaches for uncertainty analysis are presented.
Two different approaches that combine different methods are implemented accord-
ing to modelling and planning steps and domains.
First approach: Probability Random Sampling based Uncertainty Analysis
(PRSUA) use the probability approach and random sampling technique to sub sum
all uncertainty in the model output for the targeted domain, and it is presented in
intervals and quantiles. The second approach: Fuzzy Scenario based Uncertainty
Analysis (FSUA), based on the scenario approach and Fuzzy set theory is pro-
posed. The FSUA approach combines two different paradigms, probability and
Fuzzy set theory, for uncertainty analysis. The probability analysis feature in
FSUA is optional and can be implemented if expertise or a sufficient data amount
is available. The results presented by FUSA are distinct, plausible and consistent
scenarios of key drivers of a targeted domain.
The implementation of the proposed methods in case studies and evaluation
using quality factors show that:
• Both methods, particularly FSUA, outperform the existing deterministic view of
planning in IEPCT according to most quality factors.
• Both proposed multi-method approaches, FSUA and PRSUA, are resource inten-
sive and require certain expertise compared to common deterministic praxis.
However, the disadvantages in not implementing these approaches, as experience
shows, can cause incomparable expenses later.
• Compared to FSUA, the PRSUA approach is more data intensive.
• An automatic random sampling technique makes the PRSUA approach more
efficient compared to FSUA, automatizing the analysis and reducing some
subjectivity issues when experts are not available. However, as Use Case—II
shows, certain expertise will be required because of low data availability in
long-range IEPCT studies and for possible validation of models.
• The PRSUA approach can provide probabilistic information, if a sufficient
amount of information is available. However, Use Case II—Mexico shows that
data availability is very low and the results of PRSUA cannot be presented and
commented on in fully probabilistic terms. The presented probabilistic results
in Use Case II—Mexico are based partially on synthetic data to demonstrate the
implementation of PRSUA.
• FSUA use a hybrid approach combining probability and fuzzy set theory. This
makes it possible to incorporate quantitative and qualitative information, cop-
ing with linguistic uncertainty. Using probability and fuzzy set theory, FSUA
presents the results not only as fuzzy space clusters but also adds probability
information if certain expertise is available.
• Uncertainties of model output of targeted domain PRSUA presents not only par-
ametric but a different type of uncertainty accumulated in an interval with prob-
ability density.
8.5 Synthesis and conclusions of … 177

• The implementation in the Singapore Use Case shows that the planning partici-
pants’ satisfaction is higher when they have ‘natural’ scenarios using FSUA than
a probability distribution or statistical significance of results using PRSUA.
• The proposed reference impact matrix (RIM) makes it possible to deal partially
with ignorance.
Although it is too early to say which approach is the best for any IEPCT, a com-
parative assessment of the proposed approaches in previous chapter can help to
identify the appropriate approach according to the importance of the quality fac-
tors for a given IEPCT.
The PRUSA might be relevant for short planning time horizons of less than
10 years and for single sector specific planning when a sufficient data amount
is available. For long-term (several decades) integrated infrastructure planning,
FSUA will be more appropriate.
Both approaches can be implemented not only in IEP in the city or territory
level but also on the national level or in other branches such as water, transporta-
tion infrastructure or general territory planning.

8.6 Future Work

Both discussed uncertainty analysis approaches are for integrated energy planning
until an integrated plan is developed. However, there is uncertainty in planning
phase III and IV, where an integrated plan has to be fleshed out and implemented
for individual projects. Future work should be performed for uncertainty analysis
in planning phase III and IV and uncertainty types for incorporating uncertainty
perception, not only from experts but also from decision makers during the selec-
tion of options.
Uncertainty related to communication and availability of time or resources,
which are present in all planning and modelling processes, is another research
topic that needs more detailed work.
The feasibility to use possibility theory combined with the random sampling
technique in IEPCT can be another resarch area.
Appendix A
Descriptive Analysis, Modelling
of Historical Data

Different methods exist for the descriptive analysis and modelling of historical
and current situation. The analysis provide about historical development of several
model descriptors. There are different methods for descriptive analysis. The results
of some implemented methods are presented below.
Useful modelling methods are like multiplicative or an additive decomposi-
tion model. The results of additive decomposition model is shown in Fig. A.1. It is
defined as follow
xt = mt + st + zt
where, at time t, xt is the observed series, mt is the trend, st is the seasonal effect,
and zt is an error term that is, in general, a sequence of correlated random variables
with mean zero.
The graphic shows that there was significant trend of minimum temperature in
Singapore in the last decades.
Different methods exist for definition of distribution of existing data and identi-
fication of possible outliner. For interactive analysis and communication of results
a histogram presents the distribution of a univariate data set graphically, which
gives the opportunity to analyse the skewness or spread of data, identify median or
potential outliners (Fig. A.2).
However, for more precise definition of the distribution goodness of fit tests
like Kolmogorov-Smirnov or Anderson-Darling tests can be useful.
Another useful graphical tool to calculate and present low and upper quar-
tiles, median or interquartile range IQR is box plot. This gives also the opportu-
nity to identify potential outliers. However, the outliers need contextual revision
before removing or replacing from original data set. According to Tukey’s Outlier
Filter [1] only in one dataset some outliner was found across all 53 time series for
Singapore or Mexico City datasets (Fig. A.3).

© Springer International Publishing Switzerland 2015 179

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8
180 Appendix A: Descriptive Analysis, Modelling of Historical Data

Fig. A.1  Additive decomposition of min air temperature in Singapore

Appendix A: Descriptive Analysis, Modelling of Historical Data 181

Fig. A.2  Histogram, distribution of population ages 65 and above in Singapore

Fig. A.3  Single box plot, share of renewables in total energy, in Mexico

1. Hoaglin DC, Mosteller F, Tukey JW (1983) Understanding robust and explor-

atory data analysis. Wiley, New York, p 447
Appendix B
Some Empirical Results
of Use Case I-Singapore

Name: Key Factor or Driving Forces

Descriptive data analysis (historical analysis of historical data)

© Springer International Publishing Switzerland 2015 183

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8
184 Appendix B: Some Empirical Results of Use Case I-Singapore

Inductive analysis

Population Total (Inhabitants)

Descriptive data analysis

Appendix B: Some Empirical Results of Use Case I-Singapore 185

Inductive analysis

Electricity Prices (¢/kWh)

Descriptive data analysis

186 Appendix B: Some Empirical Results of Use Case I-Singapore

Inductive analysis

Population Ages 0–14 of Total (% of Total Pop.)

Descriptive data analysis

Appendix B: Some Empirical Results of Use Case I-Singapore 187

Inductive analysis

Population Ages 65 and Above (% of Total Pop.)

Descriptive data analysis

188 Appendix B: Some Empirical Results of Use Case I-Singapore

Inductive analysis
Appendix B: Some Empirical Results of Use Case I-Singapore 189

Life Style (% of Total Pop.)

Descriptive data analysis [2]

Values orientation of the seven clusters (%) Values orientation

Traditional family oriented: 16.1 Least materialistic; low pecuniary adherence;
spender; high moral standards; very pro-family
and religious
New age family: 13.9 Women should have own career; lifelong
learning is important; environmentally conscious.
concerned about moral decay; profanity
Entrepreneurs: 13.1 Like stimulation and changes; value education;
work smart rather than work hard; sense of
accomplishment is important; non-traditional;
more liberal about premarital sex and divorce
Aspirers: 18.4 Early adopters of new ideas/products; would
rather fight than compromise; materialistic; enjoy
attention; highly pro-family; observe traditions
Materialists: 14.3 Not environmentally conscious; materialistic;
financial security very important; not society
conscious; less trusting of others
Pragmatists: 11.1 Least adventurous; low entrepreneurial spirit;
dislike attention and do not worry about
social status; low respect for authority;
pro-Westernization
Independents: 13.1 Pro-feminism; low value for education; not too
concerned about financial security; liberal moral
standards; not too pro-family

Inductive analysis
“Increases in consumer spending and the trends observed in values and life-
styles are expected to continue for the coming years” p. 555
190 Appendix B: Some Empirical Results of Use Case I-Singapore

Ownership of Refrigerators (% of Total Pop.)

Descriptive data analysis

Inductive analysis
Appendix B: Some Empirical Results of Use Case I-Singapore 191

Ownership of Air Conditioner (% of Total Pop.)

Descriptive data analysis

Inductive analysis
192 Appendix B: Some Empirical Results of Use Case I-Singapore

1. Hoaglin DC, Mosteller F, Tukey JW (1983) Understanding robust and explor-

atory data analysis. Wiley, New York, p 447
2. Pecotich A, Shultz CJ (2006) Handbook of markets and economies: East Asia,
Southeast Asia, Australia. Sharpe, New Zealand, M.E
Appendix C
Some Empirical Results
of Use Case II-Mexico

Name: Key Factor or Driving Forces

Descriptive data analysis (historical analysis)

© Springer International Publishing Switzerland 2015 193

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8
194 Appendix C: Some Empirical Results of Use Case II-Mexico

Inductive analysis

Urban Population Total (Inhabitants)

Descriptive data analysis

Appendix C: Some Empirical Results of Use Case II-Mexico 195

Inductive analysis

Population Ages 0–14 of Total (% of Total Pop.)

Descriptive data analysis

196 Appendix C: Some Empirical Results of Use Case II-Mexico

Inductive analysis

Population Ages 65 and Above (% of Total Pop.)

Descriptive data analysis

Appendix C: Some Empirical Results of Use Case II-Mexico 197

Inductive analysis
Appendix D
Comparison Different Extrapolation,
Data Driven Methods and Intervals

© Springer International Publishing Switzerland 2015 199

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8
200 Appendix D: Comparison Different Extrapolation, Data Driven Methods and Intervals
Appendix D: Comparison Different Extrapolation, Data Driven Methods and Intervals 201
202 Appendix D: Comparison Different Extrapolation, Data Driven Methods and Intervals
Appendix D: Comparison Different Extrapolation, Data Driven Methods and Intervals 203
204 Appendix D: Comparison Different Extrapolation, Data Driven Methods and Intervals
Index

A E
Applied models, 10, 30 Epistemic uncertainty, 69
Error, 16, 53, 55, 69, 108, 109
Expansion, 12, 13, 63
B Expressed models, 29
Belief, 14, 17, 18, 20, 21, 98
Business planning, 2, 6
F
Flexible planning, 6, 35
C Forecasting, 7, 45, 47, 54, 55, 58, 62, 107
Calibrated models, 30 Foresight, 7, 11, 42, 43, 51, 59, 123
Case, 3, 6, 11, 14, 17, 28, 33, 41, 46, 63, 70, Formal model, 29, 30, 92
118, 147 Fuzzy-Scenario based Uncertainty Analysis
Causal forces, 42, 61, 62, 94, 146 approach (FSUA), 81, 87, 93, 94, 127,
Complementarity, 12, 13 131, 132, 167, 168, 176, 177
Complex system, 3, 5, 8, 9, 12, 25, 45, 46, 93,
97, 112
Computer models, 29 I
Conceptual model, 10, 29, 30, 105, 111, 138 Ignorance, 16, 81, 120, 167, 169
Initiation, 12, 13, 84
Initiative, 3, 5, 7, 33, 45, 70, 71, 138
D
Data-driven modelling, 44, 94, 107, 112, 165
Decision making uncertainty, 70 J
Development, 2, 4, 6, 7, 11, 13, 18, 20, 33, 39, Judgmental-driven modelling, 44, 46
40, 46, 62, 63, 71, 75, 78, 85, 86, 98,
99, 105, 108, 116, 126, 132, 138, 151,
152, 167, 170 L
Distant domain, 40–42, 45, 51, 59, 86, 93, Levels of uncertainty, 16, 33, 71
138, 147, 160, 166 Linguistic uncertainty, 16, 69, 103, 118, 144,
Domain, 5, 12, 42, 43, 45, 59, 63, 85, 89, 97, 159, 167, 169
142, 166
Driving forces, 42, 43, 60, 144

© Springer International Publishing Switzerland 2015 205

A. Mirakyan and R. De Guio, Three Domain Modelling and Uncertainty Analysis,
Energy Systems, DOI 10.1007/978-3-319-19572-8
206 Index

M S
Mental models, 10, 29 Scenario, 2, 3, 5, 16, 71, 75, 85, 115, 117, 119,
Metasystem, 40, 42 121, 148, 150, 167–169
Mixing, 12, 14, 101, 105, 112, 167 Sociotechnical system, 28, 82
Model, 3, 4, 10, 11, 28–30, 39, 40, 42, 45, 46, Strategic planning, 5, 6, 26, 136
48, 54–56, 60, 68, 69, 71, 75, 81, 82, 88, Strategy, 7, 75, 115
94, 95, 108, 111, 113, 131, 132, 137, System, 1, 3, 7, 8, 10, 28, 29, 40, 43, 45, 50,
143, 147, 157, 158, 160, 161, 164, 165 51, 69, 92, 111, 123, 132, 154

N T
Necessity, 20 Tactical planning, 26
Neighbouring domain, 42, 43, 51, 62, 86, 89, Targeted domain, 41, 42, 47, 62, 84, 92, 112,
97, 109, 114, 137, 138, 142, 147, 161, 133, 138, 147, 170
165 Theorizing, 14, 15
Timing, 14
Triangulation, 12, 13, 63, 84, 86, 94, 101, 103,
O 106, 108, 110, 112, 114, 122, 124, 126,
Operative planning, 6, 26 142, 165, 169

P U
Planning, 3–6, 10, 11, 15, 27–29, 33, 41, 70, Uncertainty, 1–4, 11, 15, 28, 32, 67–70, 79,
71, 97, 100, 101, 104, 106, 110, 115, 80, 84, 89, 96, 104, 110, 126, 138, 142,
126, 132, 164, 165, 168, 169, 172 147, 152, 157, 166
Plausibility, 17, 20, 21, 117
Policy planning, 2, 6
Possibility, 17, 19, 20, 33, 76, 98 V
Probabilistic, Random Sampling based Variability uncertainty, 70
Uncertainty Analysis approach
(PRSUA), 81, 90, 176
Procedural uncertainty, 70 W
Process-driven modelling, 44, 45 Weighting, 14
Public planning, 2, 6

R
Reference impact matrix, 59, 86, 94, 97, 138,
146, 154
Rigid planning, 6
Risk, 3, 16, 75, 76, 99, 167
Rolling planning, 6