See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/332098716

Uncertainty cost functions for solar photovoltaic generation, wind energy

generation, and plug-in electric vehicles: mathematical expected value and
verification by Monte Carlo sim...

Article  in  International Journal of Power and Energy Conversion · March 2019

DOI: 10.1504/IJPEC.2019.098620

CITATIONS READS

13 880

3 authors, including:

Juan Arévalo Bonilla Sergio Rivera

National University of Colombia National University of Colombia
2 PUBLICATIONS   24 CITATIONS    137 PUBLICATIONS   406 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Impact of renewable power plants in the energy market price View project

MONO-OBJECTIVE MINIMIZATION OF OPERATION COST FOR A MICROGRID WITH RENEWABLE POWER GENERATION, ENERGY STORAGE AND ELECTRIC VEHICLES View
project

All content following this page was uploaded by Sergio Rivera on 30 March 2019.

The user has requested enhancement of the downloaded file.

Int. J. Power and Energy Conversion, Vol. 10, No. 2, 2019 171

Uncertainty cost functions for solar photovoltaic

generation, wind energy generation, and plug-in
electric vehicles: mathematical expected value and
verification by Monte Carlo simulation

Juan Camilo Arevalo, Fabian Santos and

Sergio Rivera*
Department of Electrical and Electronics Engineering,
Electrical University
National and Electronics Engineering Deparment
of Colombia,
Universidad Nacional
Bogota, Colombia de Colombia
jucarevalobo@unal.edu.co
Email: jucarevalobo@unal.edu.co
fsantosb@unal.edu.co
Email: fsantosb@unal.edu.co
srriverar@unal.edu.co
Email: srriverar@unal.edu.co
*corresponding author
*Corresponding author

Abstract: Electrical power systems which incorporate solar or wind energy

sources, or electric vehicles, must deal with the uncertainty about the
availability of injected or demanded power. This creates uncertainty costs to be
considered in stochastic economic dispatch models. The estimation of these
costs is important for proper management of energy resources and accurate
allocation of the amount of energy available for the system. In this paper,
analytical formulas of uncertainty penalty costs are calculated, for solar and
wind energy and for electric vehicles, through a mathematical expected value
formulation. In order to get the proposed uncertainty cost functions, probability
distribution functions (PDF) of the energy primary sources are considered, that
is to say: log-normal distribution for solar irradiance PDF, Rayleigh
distribution for wind speed PDF and normal distribution for loading and
unloading behaviour PDF of electric vehicles. The analytical formulation is
verified through Monte Carlo simulations.

Keywords: wind and solar energy; electric vehicles; uncertainty cost;

economic dispatch models; mathematical modelling.

Reference to this paper should be made as follows: Arevalo, J.C., Santos, F.

and Rivera, S. (2019) ‘Uncertainty cost functions for solar photovoltaic
generation, wind energy generation, and plug-in electric vehicles: mathematical
expected value and verification by Monte Carlo simulation’, Int. J. Power and
Energy Conversion, Vol. 10, No. 2, pp.171–207.

Biographical notes: Juan Camilo Arevalo obtained his Bachelor degree in

Electrical Engineering from the National University of Colombia (UNAL). He
has worked as a Research Assistant with Professor Sergio Rivera in the
Electrical Machines Laboratory at UNAL.

Fabian Santos obtained his Bachelor degree in Electrical Engineering from the
National University of Colombia. He has worked as an Engineer Auxiliar in
CODENSA-ENEL, where he obtained experience in management and
coordination of electrical works within the grid.

Copyright © 2019 Inderscience Enterprises Ltd.

172 J.C. Arevalo et al.

Sergio Rivera is currently an Assistant Professor of Electrical Engineering at

the National University of Colombia where he specialises in smart grids. He
was a Postdoctoral Researcher at Massachusetts Institute of Technology and
Masdar Institute of Science and Technology. Prior to his postdoctoral
fellowships, he worked at General Motors as a Senior Process Engineer in the
Manufacturing Engineering Area. Meanwhile, he was a Professor of Power
Systems in the District University where he taught courses in the reliability of
power systems and electricity markets. He received his PhD from the Electric
Energy Institute at the National University of San Jan in Argentina.

1 Introduction

Photovoltaic generation (PVG), wind energy generation (WEG), and plug-in electric
vehicles (PEV) have problems of variability and uncertainty about the availability of
injected or demanded power into the grid. In the traditional economic dispatch, these
elements (PVG, WEG, PEV) may be considered as non-programmable in the operation
optimisation of a power system. However, these sources or loads of energy have a
stochastic behaviour that can be modelled with a probability distribution function (PDF),
as shown in Chang (2010), Zhao et al. (2012), Arango et al. (2017) and Surender et al.
(2015). They can be thus considered in the economic dispatch (ED).
It is well known that the source of wind power (wind speed) can be modelled
properly with a Weibull probability function (Zhao et al., 2012). The source of solar
energy (solar irradiance) depends on the geographical site where photovoltaic generation
is implemented, as shown in Chang (2010), Surender et al. (2015). For instance, there are
places where solar irradiance PDF can be modelled by beta, Weibull, or log-normal
distributions (Surender et al., 2015). Another variable that currently takes more strength
and creates uncertainty about a modern power system is the incorporation of PEV into the
network. These can be used as energy storage sources and also act as load, that is to say,
they can inject or consume power into or from the system, respectively, and this power
behaviour can be described probabilistically. Through the mathematical modelling of
PDFs, it is possible to obtain an estimate of the costs associated with the uncertainty of
these energy agents (PVG, WEG, and PEV), particularly speaking of the expected value
of an uncertainty cost function (UCF). The concept of these functions was explored first
for WEG and PEV in Zhao et al. (2012), and an application of the concept for WEG and
PVG in Surender et al. (2015). Analytical expressions for UCFs in the case of WEG and
PEV are presented in Zhao et al. (2012).
The goal of this paper is to develop a mathematical formulation of the UCFs, to test
the analytical expressions mentioned and to extend the formulation to PVG. In this way,
the paper presents a detailed mathematical formulation in order to calculate the analytical
cost of the expected penalty cost associated with the generation of electricity through
renewable energy sources: PVG and WEG, as well as the expected penalty cost for the
integration of PEVs into the power grid. The analytic calculation is verified through
Monte Carlo simulations.
Monte Carlo simulations are appropriated to simulate the variability of solar, wind
and electric vehicle resources, since these resources can be modelled by known
probability functions. For instance, in Cardell and Anderson (2010), Anderson and
Cardell (2009) and Thomopoulos (2013) and Arango et al. (2017), Monte Carlo
 Uncertainty cost functions for solar photovoltaic generation 173

simulations are used to calculate the cost impacts from the uncertainty in windfarm
output and the production costs are calculated, but not the uncertainty costs caused by the
variability of the wind resource. The stochastic behaviour of wind speed, solar irradation
and drive patters several Monte Carlo simulations have been investigated, like the
mentioned in Kamankesh et al. (2016), Ruiz-Arias et al. (2016), Kantar et al. (2016), Yao
et al. (2016) references. The novelty of the proposed approach is an analytical
formulation for the uncertainty cost. With this formulation, it is possible a deterministic
evaluation, through the expected cost to be included in an economic dispatch
(for instance in a microgrid dispatch like in Mojica-Nava et al., 2017), that considers the
probability distributions of wind speed, solar irradation and drive patters. In this way, it is
not necessary to develop Monte Carlo simulations. In order to validate the formulation,
the analytical expected cost was contrasted with the expected cost through Monte Carlo
simulations. In this paper, Monte Carlo simulations are used to simulate several scenarios
of the injected or consumed power of the mentioned primary energy sources, in each
scenario is calculated a penalty cost, and finally the expected value of the penalty cost is
obtained using the mean value of the penalty cost histogram.
This paper is organised as follows: Section 2 presents the mathematical model of the
expected UCF given by penalty costs due to underestimate and overestimate. Section 3
shows the proposed analytical development in order to obtain the UCF in the cases of
PVG, WEG, and PEV. In section 4, Monte Carlo simulations are performed in order to
obtain the UCFs using the PDF of solar irradiance, wind speed, and PEV behaviour.
Section 5 presents the analysis and discussion of the results. Finally, Section 6 presents
the conclusion. Additionally, there are two appendixes that summarise the change of
variable theorem and error function identities.

2 Mathematical description of the expected UCF

Given the uncertainty of the sources or loads (PVG, WEG, and PEV), the economic
dispatch of power systems that integrate these sources of energy must include a term in
the cost model able to consider the mentioned uncertainty. This term is composed of two
parts (Zhao et al., 2012; Surender et al., 2015):
x Penalty cost due to underestimate: penalty cost in the cost model, due to the ED
model schedules an amount of power in a generator lesser than the available power
in the source. In this case, it is a penalisation for not using all the available power
from PVG, WEG, or PEV. In terms of power, the underestimated condition is given
by:
Ws ,i  Wav ,i (1)

where Wav,i is the available actual power of the generator i, and Ws,i is the scheduled
power by the ED model to the generator i.
x Penalty cost due to overestimate: penalty cost in the cost model, due to the ED model
schedules an amount of power in a generator bigger than the available power in the
source, that is to say, power not available at a given moment. In order to supply the
load, the model requires another energy source to be turned on. In this case, it is a
174 J.C. Arevalo et al.

penalisation for use another energy source or for energy not supplied. In terms of
power, the overestimated condition is given by:
Wav ,i  Ws ,i (2)

where Wav,i is the available actual power of the generator i and Ws,i is the scheduled
power by the ED model to the generator i.
In this regard, the goal of the ED model is to properly allocate the amount of scheduled
power in order to minimise the costs due to underestimate and overestimate together with
the production cost.
The proposed UCF will be given by adding the costs due to underestimate and
overestimate:
UCF Cu ,i Ws ,i , Wav,i   Co,i Ws ,i , Wav ,i  (3)

Cu,i(Ws,i, Wav,i) and Co,i(Ws,i, Wav,i) are described in the next subsections.

2.1 Mathematical formulation in the underestimated condition

A linear function is assumed to calculate the penalty cost due to underestimate (Zhao
et al., 2012):
Cu ,i Ws ,i , Wav ,i  cu ,i Wav ,i  Ws ,i  (4)

where
cu,i is the penalty cost coefficient due to underestimate
Cu,i(Ws,i, Wav,i) is the cost function due to underestimate.
In this way, the expected penalty cost due to underestimate is given by:
Wf ,i

E ª¬Cu ,i Ws ,i , Wav ,i  º¼ ³c
Ws ,i
u ,i Wav,i  Ws ,i  ˜ fW Wav,i  ˜ dWav,i (5)

where
E[Cu,i(Ws,i, Wav,i)] is the expected value of the cost due to underestimate
fW(Wav,i) is the probability of determined available power which depends on
the PDF of the energy source (PVG, WEG, PEV)
W∞,i is the maximum power output of the generator i.

2.2 Mathematical formulation in the overestimated condition

A linear function is assumed to calculate the penalty cost due to overestimate (Zhao
et al., 2012):
Co,i Ws ,i , Wav ,i  co,i Ws ,i  Wav ,i  (6)
 Uncertainty cost functions for solar photovoltaic generation 175

where
co,i the penalty cost coefficient due to overestimate
Co,i (Ws,i, Wav,i) the cost function due to overestimate.
In this way, the expected penalty cost due to overestimate is given by:
Ws ,i

E ¬ªCo,i Ws ,i , Wav ,i  ¼º ³
Wmin,i
co,i Ws ,i  Wav,i  ˜ fW Wav ,i  ˜ dWav ,i (7)

where
E[Co,i(Ws,i, Wav,i)] is the expected value of the cost due to overestimate
fW(Wav,i) is the probability of determined available power given by the PDF of
the energy source (WEG, PVG, PEV)
Wmin,i is the minimum power output of the generator i.

3 Analytical development for UCF in the cases of PVG, WEG, and PEV

Physical quantities of solar irradiance and wind speed are not deterministic variables over
time; in many cases, they can be represented by a PDF. This probabilistic property also
applies to the behaviour of the connection and disconnection of electric vehicles to the
power grid. This section presents the UCF analytical development in the three mentioned
cases.

3.1 Photovoltaic generation case

In this paper, to obtain the mathematical expression of the UCF expected value, it is
assumed that solar irradiance behaves like a log-normal PDF. That is true in specific
geographic locations according to the research presented in Chang (2010), Zhang et al.
(2013) and Meng et al. (2013). The step-by-step original analytical development
presented in this section can be used in case of a different solar irradiance PDF, like the
presented PDFs in Chang (2010). The log-normal PDF is given by:

1 ­° > ln(G )  λ @2 ½°
fG (G ) exp ® ¾ (8)
G E 2π ¯° 2E 2 ¿°
where G is the solar irradiance, fG(G) is the corresponding log-normal PDF, λ and E are
the location and scale parameters of the log-normal distribution.
In order to determine the relationship between solar irradiance and active power
generated by a photovoltaic solar plant (WPV ), the following conditions of the energy
conversion function are defined according to the research presented in Surender et al.
(2015) depending on a reference irradiance value (Rc) according to the specific
geographical location.
x Condition A: for 0 < G = Rc
WPV follows a power quadratic transformation in relation to G:
176 J.C. Arevalo et al.

WPVr G 2
WPV (G ) (9)
Gr Rc

x Condition B: for G > Rc

WPV follows a power lineal transformation in relation to G:
WPVr G
WPV (G ) (10)
Gr
where in (9) and (10): G is the solar irradiance, Gr is the rated irradiance of the
geographical environment, WPVr is the rated active power of the PVG source. The
irradiance values are in W / m2.
In order to obtain the probability of a determined available power ( fWPV (WPV )) from the
solar irradiance PDF (fG(G)), the theory of probabilistic variable change is used. The
expression of this theorem is presented in Appendix A, and is applied for condition A and
B as follows:
x Condition A: for 0 < G = Rc
In this case, the quadratic transformation (9) and the notation g(G) function
representing the WPV power in terms of G are used.
WPVr
WPV (G ) g (G ) ˜ G2 (11)
Gr Rc
The inverse of g and its derivative are obtained:

WPV Gr Rc
g 1 WPV   (12)
WPVr

dg 1 WPV  Gr Rc 1
(13)
dWPV WPVr 2 WPV

Applying the change of variable expression (Appendix A), the WPV PDF
( fWPV (WPV )) expressed for solar photovoltaic power is obtained:

dg 1 WPV 
fWPV fG  g 1 WPV   (14)
dWPV

Then, replacing (12) (only the positive root) and (13) in (14), the PDF of the power
of the photovoltaic generator for condition A is obtained:
 Uncertainty cost functions for solar photovoltaic generation 177

ª 1 º
«§ · »
« ¨ WPV Gr Rc ¸ E 2π »
«© W PVr ¹ »
Gr Rc 1« »
fWPV WPV  (15)
WPVrWPV 2« ­ª § · º
2
½»
° «ln ¨ W G R
« ¸  λ » °»
PV r c

« ® «¬ © WPVr ¹ »¼ ¾»
«˜ exp ° °»
¬ ¯ 2E 2 ¿¼
Expression (15) is valid for the power limits of condition A:
WPV r Rc
0  WPV  (16)
Gr

x Condition B: for G > Rc

In this case, the linear transformation (10) and the notation g(G) function
representing the WPV power in terms of G are used.
WPV r
WPV (G ) g (G ) ˜G (17)
Gr
The inverse of g and its derivative are obtained:
WPV ˜ Gr
g 1 WPV  (18)
WPV r

dg 1 WPV  Gr
(19)
dWPV WPV r
Then, replacing (18) and (19) in (14), the PDF of the power of the photovoltaic
generator for condition B is obtained:

­ ª § WPV Gr · º 2 ½
° ln λ °
1 ® « ©¨ WPV r ¹¸ ¼» ¾ Gr
fWPV WPV  ˜ exp ° ¬ °˜ (20)
§ WPV ˜ Gr · ¯ 2E 2 ¿ WPV r
¨ W ¸ E 2π
© PV r ¹

The expression (20) is valid for the power limits of condition B:

WPV r Rc
WPV t (21)
Gr
Now, it is possible to obtain the penalty cost due to underestimate and overestimate
replacing (15) and (20) in (5) and (7).
178 J.C. Arevalo et al.

3.1.1 Penalty cost due to underestimate for PVG case

In order to determine the UCF part related with the penalty cost due to underestimate, the
following integral is developed:

CPV ,u ,i WPV ,i  WPV , s ,i  ˜ fWPV WPV ,i 

WPV ,f ,i

E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  º¼ ³

WPV ,s ,i
˜ dWPV ,i
(22)

where
E[CPV,u,i(WPV,s,i, WPV,i)] is the expected value of the penalty cost due to underestimate
for PVG case
fWPV (WPV ,i ) is the PDF of the power of the photovoltaic generator i

cPV,u,i is the penalty cost coefficient due to underestimate in the PVG

for generator i
WPV,∞,i is the maximum power output of the PV generator i
WPV,∞,I is the scheduled PV power set by ED model in generator i
WPV,i is the PV available power in the generator i.
The integral (22) is divided into two parts, one for condition A and another one for
condition B. A generated power is established (WRc), associated directly with the
irradiance value Rc, hence a range of generated power is assigned to the conditions A
and B:
WPV r ˜ Rc
WRc (23)
Gr

x Condition A: for 0 < WPV,i = WRc

In this case, the integral limits are WPV,s,i and WRc for equation (22). Then, it is
possible to obtain the expected penalty cost due to underestimate in condition A
(E[CPV,u,i(WPV,s,i, WPV,i), A]), replacing (15) in (22) with the mentioned integral limits
as follows:

Gr Rc 1
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , Aº¼ cPV ,u ,i
WPVr Gr Rc
2 E 2π
WPVr
WRc
WPV ,i  WPV , s ,i
˜ ³
WPV ,s ,i
WPV ,i
(24)

­ ª § · º
2
½
°° « ln ¨ WPV ,i ˜ Gr ˜ Rc ¸¸  λ » °°
® « ¨© WPVr ¹ ¼» ¾
˜ exp ° ¬ ° dWPV ,i
°¯ 2E 2 °¿
 Uncertainty cost functions for solar photovoltaic generation 179

WRc
1 WPV ,i  WPV , s ,i
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , Aº¼ cPV ,u ,i
2 E 2π
˜ ³
WPV ,s ,i
WPV ,i

­ ª § (25)
· º ½
2
° « ln ¨ WPV ,i ˜ Gr ˜ Rc ¸  λ » °
® « © WPVr ¹ »¼ ¾
˜ exp ° ¬ ° dWPV ,i
¯ 2E 2
¿
It is possible to use the variable change theorem for integrals as follows:

ª § WPV ,i ˜ Gr ˜ Rc · º
«ln ¨ ¸  λ»
¬« © WPVr ¹ ¼»
U (26)
2E

1
dU dWPV ,i (27)
2 2 E WPV ,i

From (26), it is possible to obtain WPV,i in function of U:

WPVr 2 2EU 2 λ
WPV ,i (U ) e e (28)
Gr Rc
The integral limits for the variable change integral can be defined as:
§ 1 § WPV , s ,i Gr Rc · ·
¨ ln ¨ ¸  λ¸
if WPV ,i WPV , s ,i o U a © 2 © WPVr ¹ ¹
(29)
2E

§ 1 § WRc Gr Rc · ·
¨ 2 ln ¨ W ¸  λ¸
© © PVr ¹ ¹
if WPV ,i WRc o U b (30)
2E
The integral in (25) now is:
WPVr 2 2EU 2 λ
Ub e e  WPV , s ,i
cPV ,u ,i Gr ˜ Rc
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , Aº¼
2 E 2π
˜ ³
Ua
WPV ,i (31)

˜ e U 2 2 E WPV ,i ˜ dU
2

Ub
(1)cPV ,u ,iWPV , s ,i
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , Aº¼
π
˜ ³e
Ua
U 2
˜ dU

Ub
(32)
CPV ,u ,i e2 λ  2 E WPVr
2

³e  U  2 
2
 ˜ ˜ dU
πGr Rc Ua

Solving the integrals through the error function identities of Appendix B, the
expected value of the cost function in terms of the limits Ua and Ub is obtained:
180 J.C. Arevalo et al.

cPV ,u ,iWPV , s ,i
E ¬ªCPV ,u ,i WPV , s ,i , WPV ,i  , A¼º >erf U a   erf U b @
2
CPV ,u ,iWPVr ˜ e 2 λ  2 E ªerf U b  2E  º
2 (33)
 « »
2Gr Rc «¬ erf U a  2E  »¼

Returning to the original limits (29) and (30), it is possible to obtain the following
expression for the expected penalty cost due to underestimate in condition A:
ª § § 1 § WRc Gr Rc · · · º
« ¨ ¨ ln ¨ ¸  λ¸ ¸ »
«erf ¨ © 2 © WPVr ¹ ¹ ¸ »
(1)cPV ,u ,iWPV , s ,i « ¨© 2E ¸
¹ »
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , Aº¼ « »
2 « § § 1 § WPV , s ,i Gr Rc · · · »
« ¨ ¨ ln ¨ ¸  λ ¸ ¸»
« erf ¨ © 2 © WPVr ¹ ¹ ¸»
« ¨ 2 E ¸»
¬ © ¹¼

ª § § 1 § WRc Gr Rc · · · º
« ¨ ¨ ln ¨ ¸  λ¸ ¸ »
« erf ¨ © 2 © WPVr ¹ ¹ ¸ »
« ¨ 2E ¸ »
« ¨ ¸ »
CPV ,u ,iWPVr ˜ e 2 λ  2 E
2
« ¨©  2E ¸
¹ »
 « » (34)
2Gr Rc « § § 1 § WPV , s ,i Gr Rc · · · »
« ¨ ¨ ln ¨
2 © WPVr ¸  λ ¸ ¸»
« erf ¨ © ¹ ¹ ¸»
« ¨ 2E ¸»
« ¨ ¸»
«¬ ¨  2E ¸»
© ¹¼
x Condition B: for WPV,i > WRc
In this case, the integral limits are WRc and WPV,∞,i for equation (22). Then, it is
possible to obtain the expected penalty cost due to underestimate in condition B
(E[CPV,u,i(WPV,s,i, WPV,i), B]), replacing (20) in (22) with the mentioned integral limits
as follows:
WPV ,f ,i
cPV ,u ,i WPV ,i  WPV , s ,i
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , B º¼
E 2π
˜ ³
WRc
WPV ,i

­ ª § WPV ,i ˜ Gr (35)
· º ½
2
° «ln ¨ ¸ » ¾°
 λ
® © WPVr ¹ ¼
˜ exp ° ¬ ° dWPV ,i
¯ 2E 2 ¿
It is possible to use the variable change theorem for integrals as follows:
ª § WPV ,i ˜ Gr · º
« ln ¨ W ¸  λ»
U ¬ © PVr ¹ ¼
(36)
2E
 Uncertainty cost functions for solar photovoltaic generation 181

1
dU dWPV ,i (37)
2 E WPV ,i

From (36), it is possible to obtain WPV,i in function of U:

WPVr 2EU  λ
WPV ,i (U ) e (38)
Gr
The integral limits for the variable change integral can be defined as:

§ § WRc Gr · ·
¨ ln ¨ W ¸  λ¸
if WPV ,i WRc o U a © © PVr ¹ ¹
(39)
2E

§ § WPV ,f,i Gr · ·
¨ ln ¨ ¸  λ¸
if WPV ,i WPV ,f ,i o U b © © WPVr ¹ ¹
(40)
2E
The integral in (35) now is:

§ WPVr λ U | 2E ·
Ub ¨ G e  WPV , s ,i ¸
cPV ,u ,i © r ¹
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , B º¼
E 2π U a ³ WPV ,i (41)

˜e U 2 E WPV ,i ˜ dU
2

Ub
(1)cPV ,u ,iWPV , s ,i
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , B º¼
π
˜ ³e
Ua
U 2
˜ dU

2
(42)
CPV ,u ,iWPVr e λ   E 2
2 Ub § E ·
¨ U 

³
¸
 ˜ e © 2¹ ˜ dU
πGr Ua

Solving the integrals through the error function identities of Appendix B, the
expected value of the cost function in terms of the limits Ua and Ub is obtained:
cPV ,u ,iWPV , s ,i
E ª¬CPV ,u ,i WPV , s ,i , WPV ,i  , B º¼ >erf U a   erf U b @
2
ª § E · º
« erf ¨ U b  ¸ » (43)
CPV ,u ,iWPVr ˜ e λ  E 2 « ©
2
2¹ »

2 ˜ Gr « § E ·»
« erf ¨ U a  ¸»
¬ © 2 ¹¼

Returning to the original limits (39), (40), it is possible to obtain the following
expression for the expected penalty cost due to underestimate in condition B:
182 J.C. Arevalo et al.

ª § § § WRc Gr · · · º
« ¨ ¨ ln ¨ ¸  λ¸ ¸ »
« erf ¨ © © WPVr ¹ ¹ ¸ »
cPV ,u ,iWPV , s ,i « ¨© 2E ¸
¹ »
E ¬ªCPV ,u ,i WPV , s ,i , WPV ,i  , B ¼º « »
2 « § § § WPV ,f ,i Gr · · · »
« ¨ ¨ ln ¨ ¸  λ ¸ ¸»
« erf ¨ © © WPVr ¹ ¹ ¸ »
« ¨ 2E ¸»
¬ © ¹¼

ª § § § WPV ,f ,i Gr · · · º
« ¨ ¨ ln ¨ ¸  λ ¸ ¸»
« ¨ © © WPVr ¹ ¹ ¸ »
« erf ¨ 2E ¸»
« ¨ ¸»
« ¨ E ¸»
« ¨  ¸»
CPV ,u ,iWPVr ˜ e λ  E
2

« © ¹»
2
2
 (44)
2 ˜ Gr « § § § WRc Gr · · · »
« ¨ ¨ ln ¨ ¸  λ¸ ¸ »
« ¨ © © WPVr ¹ ¹ ¸ »
« erf ¨ 2E ¸ »
« ¨ ¸ »
« ¨ E ¸ »
« ¨ ¸ »
¬ © 2 ¹ ¼

3.1.2 Penalty cost due to overestimate for PVG case

In order to determine the UCF part related with the penalty cost due to overestimate, the
following integral is developed:
WPV ,s ,i

E ª¬CPV ,o,i WPV , s ,i , WPV ,i  º¼ ³0

cPV ,o,i WPV , s ,i  WPV ,i  fWPV WPV  ˜ dWPV ,i (45)

where
E[CPV,o,i(WPV,s,i, WPV,i)] is the expected value of the penalty cost due to overestimate for
PVG case
fWPV (WPV ,i ) is the PDF of the power of the photovoltaic generator i

cPV,o,i is the penalty cost coefficient due to overestimate in the PVG

for generator i
WPV,∞,i is the maximum power output of the PV generator i
WPV,s,i is the scheduled PV power set by ED model in generator i
WPV,i is the PV available power in the generator i.
The integral (45) is divided into two parts, one for condition A and another one for
condition B. The range of generated power for conditions A and B is similar to the used in
Section 3.1.1.
 Uncertainty cost functions for solar photovoltaic generation 183

x Condition A: for 0 < WPV,i = WRc

In this case, the integral limits are 0 and WRc for equation (45). Then, it is possible to
obtain the expected penalty cost due to overestimate in the condition A
(E[CPV,o,i(WPV,s,i, WPV,i), A]), replacing (15) in (45) with the mentioned integral limits
as follows:

Gr Rc 1
E ª¬CPV ,o,i WPV , s ,i , WPV ,i  , Aº¼ cPV ,o,i
WPVr GR
2 r c E 2π
WPVr
WRc
WPV , s ,i  WPV ,i
˜ ³
0
WPV ,i
(46)

­ ª § · º
2
½
°° «ln ¨ WPV ,i ˜ Gr ˜ Rc ¸¸  λ » °°
® « ©¨ WPVr ¹ ¼» ¾
˜ exp ° ¬ ° dWPV ,i
°¯ 2E 2 °¿
WRc
1 WPV ,i  WPV , s ,i
E ª¬CPV ,o,i WPV , s ,i , WPV ,i  , Aº¼ cPV , o,i
2 E 2π
˜ ³
0
WPV ,i
­ ª § (47)
· º ½
2
° «ln ¨ WPV ,i ˜ Gr ˜ Rc ¸  λ » °
® « © WPVr ¹ »¼ ¾
˜ exp ° ¬ ° dWPV ,i
¯ 2E 2
¿
It is possible to use the variable change theorem for integrals, like in the
underestimated case. Equations (26), (27) and (28) are thus used to solve (47). The
integral limits for the variable change integral are different and can be defined as:

§ 1 § 0 ˜ Gr Rc · ·
¨ 2 ln ¨ W ¸  λ¸
© © PVr ¹ ¹
if WPV ,i 0 o Ua f (48)
2E

§ 1 § WRc Gr Rc · ·
¨ 2 ln ¨ W ¸  λ¸
© © PVr ¹ ¹
if WPV ,i WRc o U b (49)
2E
It is possible to use (33) and replace it in equations (48) and (49). The expression for
the expected penalty cost due to overestimate in condition A is then:
184 J.C. Arevalo et al.

ª § § 1 § WRc Gr Rc · · · º
« ¨ ¨ ln ¨ ¸  λ ¸ ¸»
E ª¬CPV ,o,i WPV , s ,i , WPV ,i  , Aº¼
cPV ,o,iWPV , s ,i «erf (f)  erf ¨ © 2 © WPVr ¹ ¹ ¸ »
2 « ¨ 2E ¸»
¬ © ¹¼
ª § § 1 § WRc Gr Rc · · ·º (50)
« ¨ ¨ ln ¨ ¸  λ ¸ ¸»
«erf ¨ © 2 © WPVr ¹ ¹ ¸»
CPV ,o,iWPVr ˜ e2 λ  2 E
2

 « ¨ 2E ¸»
Gr Rc 2 « ¨ ¸»
¨
« ©  2E ¸»
¹
« »
¬ erf  f  2 E 
« »¼

Returning to the original limits (48) and (49), it is possible to obtain the following
expression for the expected penalty cost due to overestimate in condition A:
ª § § 1 § WRc Gr Rc · · · º
« ¨ ¨ ln ¨ ¸  λ ¸ ¸»
cPV ,o ,iWPV , s ,i «1  erf ¨ © 2 © WPVr ¹ ¹ ¸ »
E ª¬CPV ,o ,i WPV , s ,i , WPV ,i  , Aº¼ ¨ ¸»
2 « © 2E ¹¼ (51)
¬
ª § § 1 § WRc Gr Rc · · · º
2 « ¨ ¨ ln ¨ ¸  λ¸ ¸ »
CPV ,o,iWPVr ˜ e 2 λ  2 E « ¨ © 2 © WPVr ¹ ¹
 erf ¨  2 E ¸¸  1»
Gr Rc 2 « © 2E ¹ »¼
¬

x Condition B: for WPV,i > WRc

In this case, the integral limits are WRc and WPV,s,i for equation (45). Then, it is
possible to obtain the expected penalty cost due to overestimate in condition B
(E[CPV,o,i(WPV,s,i, WPV,i), B]), replacing (20) in (45) with the mentioned integral limits
as follows:
WPV ,s ,i
cPV ,o,i WPV ,i  WPV , s ,i
E ¬ªCPV ,o,i WPV , s ,i , WPV ,i  , B ¼º
E 2π
˜ ³
WRc
WPV ,i

­ ª § (52)
· º ½
2
° «ln ¨ WPV ,i ˜ Gr ¸  λ » °
® « © WPVr ¹ »¼ ¾
˜ exp ° ¬ ° ˜ dWPV ,i
¯ 2E 2 ¿
It is possible to use the variable change theorem for integrals, like in the
underestimated case and condition A. Equations (36), (37), and (38) are thus used to
solve (52). The integral limits for the variable change integral are different and can
be defined as:

§ § WRc Gr · ·
¨ ln ¨ W ¸  λ¸
if WPV ,i WRc o U a © © PVr ¹ ¹ (53)
2E

§ § WPV , s ,i Gr · ·
¨ ln ¨ ¸  λ¸
if WPV ,i WPV , s ,i o U a © © WPVr ¹ ¹ (54)
2E
 Uncertainty cost functions for solar photovoltaic generation 185

It is possible to use (43) and replace it in equations (53) and (54). The expression for
the expected penalty cost due to overestimate in condition B is then:
ª § § § WRc Gr · · · º
« ¨ ¨ ln ¨ ¸  λ¸ ¸ »
«erf ¨ © © WPVr ¹ ¹ ¸ »
cPV ,o ,iWPV , s ,i « ¨© 2E ¸
¹ »
E ª¬CPV ,o ,i WPV , s ,i , WPV ,i  , B º¼ « »
2 « § § § WPV , s ,i Gr · · · »
« ln
¨¨ ¨  λ
¸ ¸ ¸»
« erf ¨ © © WPVr ¹ ¹ ¸ »
« ¨ ¸» (55)
¬ © 2E ¹¼
ª § § § WPV , s ,i Gr · · ·º
« ¨ ¨ ln ¨ ¸  λ¸ ¸»
«erf ¨ © © WPVr ¹ ¹  E ¸ »
CPV ,o,iWPVr ˜ e λ  E
2 2 « ¨ 2E 2 ¸¹ »
 « © »
2 ˜ Gr « § § § WRc Gr · · · »
« ¨ ¨ ln ¨ ¸  λ¸ ¸ »
« erf ¨ © © WPVr ¹ ¹  E ¸ »
« ¨ 2E 2 ¸¹ »¼
¬ ©

In this way, it is possible to get the UCF for PVG case by adding the equations (34),
(44), (51) and (55).

3.2 Wind power generation case

Wind speed (υ), as a primary source of energy for WEG, can be considered with a
behaviour according to a Rayleigh PDF (56) in several worldwide locations (Chang,
2010; Zhao et al., 2012; Surender et al., 2015).

υ ­° § υ ·2 ½°
fυ (υ) exp ® ¨ ¸ ¾ (56)
σ2 ¯° © 2σ ¹ °¿
In (56), the Rayleigh scale parameter is σ. In order to determine the relationship between
wind speed and active power generated by a WEG (Ww), the following conditions of the
energy conversion function are defined according to the research presented in Zhao et al.
(2012), this depends on a cut-in wind speed (vi), a cut-out wind speed (vo), a rated wind
speed (vr), a rated power output (Wr), and ρ and κ constants given by:
Wr
ρ (57)
 υr  υi 
Wr ˜ υi
κ (58)
 υr  υi 
x Condition A: for υ ≤ υi or υ ≥ υo
In this case, the power generated due to insufficient wind energy or saturation in the
generator is 0.
if υ d υi or υ t υ0 then Ww (υ) 0 (59)
186 J.C. Arevalo et al.

The probability of the wind power output being zero coincides with the sum of the
probability of wind speed being smaller than υi plus the probability of wind speed
being larger than υo:
§ °­ § υ · °½
2
°­ § υ · °½ ·
2
fW Ww 0 ¨1  exp ® ¨ i ¸ ¾  exp ® ¨ o ¸ ¾ ¸ (60)
¨ °¯ © 2σ ¹ °¿ °¯ © 2σ ¹ °¿ ¸¹
©
x Condition B: for υi < υ < υr
In this case, there is a linear wind speed ratio between υ and power Ww.
if υi  υ  υr t υ0 then Ww (υ) ρυ  κ (61)

Based on the wind speed PDF (56), the following PDF is obtained for wind power
Ww values in the range of wind speed of condition B (considering the variable change
properties, since the transformation from variable υ to variable Ww is used, where the
relationship Ww(υ) is known):

§ Ww  κ · ­° § Ww  κ ·2 ½°
fW  0  Ww  Wr  ¨ ρ2σ 2 ¸ exp ® ¨ ¸ ¾ (62)
© ¹ ¯° © 2 ρσ ¹ ¿°
x Condition C: for υr ≤ υ < υo
In this case, there is a constant power output with respect to the wind speed.
if υr d υ  υo then Ww (υ) Wr (63)

Based on the wind PDF (56), the following PDF is obtained for wind power Ww
values in the range of wind speed condition C, using the probability of wind speed
between υr and υo:
§ ­° § υ ·2 ½° ­° § υ ·2 ½° ·
fW Ww Wr  ¨ exp ® ¨ r ¸ ¾  exp ® ¨ o ¸ ¾ ¸ (64)
¨ ¸
© ¯° © 2σ ¹ ¿° ¯° © 2σ ¹ ¿° ¹

3.2.1 Penalty cost due to underestimate for WEG case

In order to determine the UCF part related with the penalty cost due to underestimate, the
following integral is developed:
Wr

E ª¬Cw,u ,i Ww, s ,i , Ww,i  º¼ ³

Ww ,s ,i
cw,u ,i Ww,i  Ww, s ,i  ˜ fW Ww,i  dWw,i (65)

where
E[Cw,u,i(Ww,s,i, Ww,i)] is the expected value of the penalty cost due to underestimate
for WEG case
fW(Ww,i) is the PDF of the power of the WEG generator i
cw,u,i is the penalty cost coefficient due to underestimate in the WEG
for generator i
 Uncertainty cost functions for solar photovoltaic generation 187

Wr is the maximum power output of the WEG generator i

Ww,s,i is the scheduled WEG power set by ED model in generator i
Ww,i is the WEG available power in the generator i.
The integral (65) is divided into two parts, one for condition B and another for
condition C:

Wr
cw,u ,i Ww,i  Ww, s ,i 
E ª¬Cw,u ,i Ww, s ,i , Ww,i  º¼ ³
Ww ,s ,i ˜
Ww,i  κ ¨©
e
§ Ww ,i  κ ·
2 ρσ ¹
¸
δ Ww,i  Wr  dWw,i
ρ2 σ 2 (66)


Wr

³c w, u , i Ww,i  Ww, s ,i  ˜  § υ2 ·
¨ r ¸
e © 2σ 2 ¹
§ υ2 ·
¨ 0 ¸
 e © 2σ 2 ¹  dW
w, i
Wr

In order to solve the above integral (66), the following identities are used:
x
2
π³
erf ( x) e t 2
˜ dt (67)
0

­ f  x0  si a  x0  b
b

³ f ( x)δ  x  x  ˜ dx
a
0 ®
¯ 0 si  a; x0 ! b
(68)

b
1   a2
³ te t 2
˜ dt e  eb 
2
(69)
2
a

b
1 1
³t e2 t 2
˜ dt π  erf (b)  erf ( a)    ae a   beb 
2 2
(70)
4 2
a

The integral (65) was divided into two parts, the first part is denominated int1:
2
Wr § Ww ,i  κ ·
Ww,i  κ ¨ ¸
int1 ³
Ww ,s ,i
cw, s ,i Ww,i  Ww, s ,i  ˜ 2 2 e ©
ρ σ
2 ρσ ¹
dWw,i (71)

It is possible to use the variable change theorem for integrals as follows:

Ww,i  κ
U o Ww,i 2 ρσU  κ (72)
2 ρσ

dWw,i
dU o dWw,i 2 ρσ ˜ dU (73)
2 ρσ

The integral limits for the variable change integral can be defined as:
Ww, s ,i  κ
if Ww,i Ww, s ,i o U a (74)
2 ρσ
188 J.C. Arevalo et al.

Wr  κ
if Ww,i Wr o U b (75)
2 ρσ

In this way, it is possible to develop (71):

Ub

³c  2 ρσU  κ  Ww, s ,i  ˜
2U U 2
int1 w, u , i e 2 ρσdU (76)
ρσ
Ua

Ub

³ 2 ρσU  κ  Ww, s ,i  ˜ Ue U dU
2
int1 2cw,u ,i (77)
Ua

ª Ub Ub
º
³
2cw,u ,i « 2 ρσ Ue U dU   κ  Ww, s ,i  Ue U ˜ dU » ³
2 2
int1 (78)
«¬ Ua Ua
»¼

Using the identities of the equations (69) and (70), (79) and (80) are obtained:

cw,u ,i ª 2π ρσ  erf U a   erf U b    2 ρσ  2U a e U a  2U b e b  º

2 U 2
int1 « » (79)
2 « 2  κ  Ww, s ,i   e U a2  e U b2  »
¬ ¼

cw , u , i 
ª 2π ρσ  erf U a   erf U b    e U a2 2 2 ρσU a  2  κ  Ww, s ,i 
«
º»
int1 (80)
2
¬ 
« e U b2 2 2 ρσU b  2  κ  Ww, s ,i   »
¼
Solving the following terms:
Ww, s ,i  κ
2 2 ρσU a  2  κ  Ww, s ,i  2 2 ρσ  2  κ  Ww, s ,i  0 (81)
2 ρσ

Wr  κ
2 2 ρσU b  2  κ  Ww, s ,i  2 2 ρσ  2  κ  Ww, s ,i  2 Wr  Ww, s ,i  (82)
2 ρσ

Then, integral (71) is:

cw , u , i ª
2π ρσ  erf U a   erf U b    2 Ww, s ,i  Wr  e U b º¼
2
int1 ¬ (83)
2
The second part of the integral is denominated int2:

int2
Wr

³c w, u , i Ww,i  Ww, s ,i  ˜ e
§ υ2 ·
¨ r 2 ¸
© 2σ ¹
§ υ2 ·
¨ 02 ¸
 e © 2σ ¹ δ Ww,i  Wr  dWw,i (84)
Wr

Developing the integral, the following expression is obtained:

int2 e  § V2 ·
cw,u ,i ¨© 2σr 2 ¸¹ ¨© 2σ0 2 ¸¹
e
§ V2 ·

Wr  Ww, s,i  (85)
2
 Uncertainty cost functions for solar photovoltaic generation 189

In this way, the expected cost due to underestimate is:

E ª¬Cw,u ,i Ww, s ,i , Ww,i  º¼

cw , u , i
2
 2π ρσ  erf U a   erf U b   

2 Ww, s ,i , Wr   e U b
2
(86)


cw , u , i
e  V2
 r2
2σ e
V2
 02
2σ  W  W
r w, s ,i 
2
Returning to the original limits given by the variable change [(74) and (75)], it is possible
to obtain the following expression for the expected penalty cost due to underestimate:
§ § § Ww, s ,i  κ · § Wr  κ · · ·
¨ 2π ρσ ¨ erf ¨ ¸  erf ¨ ¸¸¸
cw , u , i ¨ © © 2 ρσ ¹ © 2 ρσ ¹ ¹ ¸
E ¬ªCw,u ,i Ww, s ,i , Ww,i  ¼º
2 ¨ § W κ ·
2 ¸
¨ ¨ r ¸ ¸ (87)
¨ 2 Ww, s ,i , Wr   e © 2 ρσ ¹ ¸
© ¹

 e 
cw,u ,i  2Vσr 2  0
2

 e 2σ 2
V2
 W  W
r w, s , i 
2

3.2.2 Penalty cost due to overestimate for WEG case

In order to determine the UCF part related with the penalty cost due to overestimate, the
following integral is developed:
Ww ,s ,i

E ª¬Cw,o,i Ww, s ,i , Ww,i  º¼ ³

0
cw,o,i Ww, s ,i  Ww,i  fW Ww,i  dWw,i (88)

where
E[Cw,o,i(Ww,s,i, Ww,i)] is the expected value of the penalty cost due to overestimate for
WEG case
fW(Ww,i) is the PDF of the power of the WEG generator i
cw,o,i is the penalty cost coefficient due to overestimate in the WEG
for generator i
Ww,s,i is the scheduled WEG power set by ED model in generator i
Ww,i is the WEG available power in the generator i.
The integral (88) is divided into two parts, one for condition A and another for
condition B:
0

E ª¬Cw,o,i Ww, s ,i , Ww,i  º¼ ³c

0
w, o , i Ww, s ,i  Ww,i  fW (0)δ Ww,i  dWw,i
Ww ,s ,i
(89)

0
³ cw,o,i Ww, s ,i  Ww,i  fW Ww,i  dWw,i
190 J.C. Arevalo et al.

E ª¬Cw,o,i Ww, s ,i , Ww,i  º¼

0

³c w, o ,i Ww, s ,i  Ww,i  ˜ 1  e

υr2
2σ 2 e

υ02
2σ 2  dW w, i
0
2
(90)
Ww ,s ,i § Ww ,i  κ ·
Ww,i  κ ¨ ¸

0
³ cw,o,i Ww, s ,i  Ww,i  ˜ 2 2 e ©
ρ σ
2 ρσ ¹
dWw,i

The integral (88) was divided into two parts, the first part is denominated int1:
2
Ww ,s ,i § Ww ,i  κ ·
Ww,i  κ ¨ ¸
int1 ³
0
cw,u ,i Ww,i  Ww, s ,i  ˜ 2 2 e ©
ρ σ
2 ρσ ¹
dWw,i (91)

It is possible to use the variable change theorem for integrals as follows:

Ww,i  κ
U o Ww,i 2 ρσU  κ (92)
2 ρσ

dWw,i
dU o dWw,i 2 ρσ ˜ dU (93)
2 ρσ

The integral limits for the variable change integral can be defined as:
Ww, s ,i  κ
if Ww,i Ww, s ,i o U b (94)
2 ρσ

κ
if Ww,i 0 o U a (95)
2 ρσ

In this way, it is possible to develop (91):

Ub

³ c W  2U U 2
int1 w, o , i w, s , i   2 ρσU  κ  ˜ e 2 ρσdU (96)
ρσ
Ua

Ub

³ c W  2U U 2
int1 w, o , i w, s , i   2 ρσU  κ  ˜ e 2 ρσdU (97)
ρσ
Ua

Ub Ub

³
2cw,o,i Ww, s ,i  κ  Ue U dU  2 2cw,o,i ρσ U 2 e U dU ³
2 2
int1 (98)
Ua Ua

Using the identities of equations (69) and (70), (99) and (100) are obtained:
1  U a2
2cw,o,i Ww, s ,i  κ   e U b 
2
int1 e
2
(99)
§1 2 ·
π  erf U b   erf U a    U a e U a  U b eU b  ¸
1
2 2 ρσcw,o,i ¨
2

©4 2 ¹
 Uncertainty cost functions for solar photovoltaic generation 191

cw,o,i Ww, s ,i  κ   eU a  e U b 

2 2
int1
§ κ U a2 Ww, s ,i  κ U b2 ·
 2 ρσcw,o,i ¨ e  e ¸ (100)
© 2 ρσ 2 ρσ ¹
2πcw,o,i ρσ
  erf U b   erf U a  
2
Then, the integral (91) is:

int1
2

cw,o,i Ww, s ,i  κ   eU a  e U b   cw,o,i κe U a  Ww, s ,i  κ  e U b
2 2 2

(101)
2πcw,o,i ρσ
  erf U b   erf U a  
2

2πcw,o,i ρσ
int1 cw,o,iWw, s ,i e U a 
2
 erf U b   erf U a   (102)
2
The second part of the integral is denominated int2:

int2
0

³ cw,o,i Ww, s ,i  Ww,i  ˜ 1  e  

Vi2
2σ 2 e

V02
2σ 2 δ W w, i  dWw,i (103)
0

Developing the integral, the following expression is obtained:

int2 
cw,o,iWw, s ,i ˜ 1  e

Vi2
2σ 2 e

V02
2σ 2  (104)

In this way, the expected cost due to overestimate is:

E ª¬Cw,o,i Ww, s ,i , Ww,i  º¼ cw,o,iWw, s ,i ˜ 1  e


Vi2
2σ 2 e

V02
2σ 2 e

κ2
2 ρ2 σ 2 
(105)
2πcw,o,i ρσ
  erf U b   erf U a  
2
Returning to the original limits given by the variable change [(94) and (95)], it is possible
to obtain the following expression for the expected penalty cost due to overestimate:

E ¬ªCw,o,i Ww, s ,i , Ww,i  ¼º cw,o,iWw, s ,i ˜ 1  e


Vi2
2σ 2 e

V02
2σ 2 e

κ2
2 ρ2 σ 2 
(106)
2πcw,o,i ρσ § § Ww, s ,i  κ · § κ · ·
 ¨ erf ¨ ¸  erf ¨ ¸¸
2 © © 2 ρσ ¹ © 2 ρσ ¹ ¹

In this way, it is possible to get the UCF for WEG case by adding the equations (87) and
(106).
192 J.C. Arevalo et al.

3.3 PEV case

According to several simulation studies, PEVs have a loading and unloading behaviour
that can be modelled with a normal PDF (Zhao et al., 2012; Guo et al., 2012; Xie et al.,
2011; Sufen et al., 2012):
2
§ P μ ·
1 ¨ e ¸
f Pe  Pe  e © 2I ¹ (107)
2πI 2

where f Pe is the PDF of the power in the batteries of PEVs, Pe represents the batteries’
available power, µ and I are the mean and standard deviation respectively of the PEV
PDF.

3.3.1 Penalty cost due to underestimate for PEVs case

In order to determine the UCF part related with the penalty cost due to underestimate, the
following integral is developed:
Pe ,f

E ª¬Ce,u ,i  Pe,i , Pe, s ,i  º¼ ³c

Pe ,s ,i
e,u ,i  Pe,i  Pe, s ,i  ˜ f P  Pe,i  ˜ dPe,i
e ,i (108)

where
E[Ce,u,i(Pe,i, Pe,s,i)] is the expected value of the penalty cost due to underestimate for
PEV case
f Pe  Pe,i  is the PDF of the power available of PEV in node i

Ce,u,i is the penalty cost coefficient due to underestimate in the PEV in

node i
Pe,∞ is the maximum injected power of PEV in node i
Pe,s,i is the scheduled PEVs power set by ED model in node i
Pe,i is the available PEV power in the node i.
Replacing equation (107) in (108), (109) is found:
2
Pe ,f § Pe ,i  μ ·
1 ¨ ¸
E ª¬Ce,u ,i  Pe,i , Pe, s ,i  º¼ ³c
Pe ,s ,i
e,u ,i  Pe,i  Pe, s ,i  ˜
2πI
e © 2I ¹ dPe,i (109)

It is possible to use the variable change theorem for integrals as follows:

Pe,i  μ
U o Pe,i U 2I  μ (110)
2I

dPe,i
dU o dPe,i dU 2I (111)
2I
 Uncertainty cost functions for solar photovoltaic generation 193

The integral limits for the variable change integral can be defined as:
Pe, s ,i  μ
if Pe,i Pe, s ,i o U a (112)
2I

f  μ
if Pe,i f o U b tend to f  (113)
2I
In this way, it is possible to develop (109):
Ub

³ c U 2I  μ  Pe, s ,i  ˜
1
E ¬ªCe,u ,i  Pe,i , Pe, s ,i  ¼º eU ˜ 2I dU
2
e,u ,i
Ua
2πI
(114)
 μ  Pe, s ,i 
Ub Ub
2I
Ua
³c e ,u ,i
π
˜ Ue U 2
dU  ³c
Ua
e ,u , i
π
˜e U 2
dU

Using the identities (69) and (138), the following expression is obtained:
ce,u ,i
E ª¬Ce,u ,i  Pe,i , Pe, s ,i  º¼  μ  Pe, s ,i   erf U b   erf U a  
2
(115)
ce,u ,i 2 1  U a2
 eU b 
2
 ˜ e
π 2

Returning to the original limits given by the variable change [(112) and (113)], it is
possible to obtain the following expression for the expected penalty cost due to
underestimate:
§ P μ ·
 μ  Pe, s ,i  ¨ erf  f    erf §¨ e, s ,i ·¸ ¸
ce,u ,i
E ª¬Ce,u ,i  Pe,i , Pe, s ,i  º¼
2 © © 2I ¹ ¹
(116)
§ § Pe ,s ,i  μ ·2 ·
ce,u ,iI ¨ ¨© 2I ¸¹  ¸
 ˜©e  ef ¹
2π
It is known that erf(∞+) = 1, and that f(x) = erf(x) is an odd function; the final expression
of the expected penalty cost due to underestimate is achieved by:
§ μ  Pe, s ,i ··
 μ  Pe, s ,i  ¨1  erf §¨
ce,u ,i
E ¬ªCe,u ,i  Pe,i , Pe, s ,i  ¼º ¸¸
2 © © 2I ¹¹
2
(117)
§ μ  Pe ,s ,i ·
ce,u ,i ˜ I ¨ ¸
 ˜ e © 2I ¹
2π
194 J.C. Arevalo et al.

3.3.2 Penalty cost due to overestimate for PEV case

In order to determine the UCF part related with the penalty cost due to overestimate, the
following integral is developed:
Pe ,s ,i

E ª¬Ce,o,i  Pe,i , Pe, s ,i  º¼ ³c

0
e, o ,i  Pe, s ,i  Pe,i  ˜ f p  Pe,i  ˜ dPe,i
e (118)

where
E[Ce,o,i(Pe,i, Pe,s,i)] is the expected value of the penalty cost due to overestimate for
PEV case
f Pe  Pe,i  is the PDF of the power available of PEV in node i

ce,o,i is the penalty cost coefficient due to overestimate in the PEV in

node i
Pe,s,i is the scheduled PEV power set by ED model in node i
Pe,i is the available PEV power in the node i.
Replacing equation (107) in (118), (119) is found:
2
Pe ,s ,i § Pe ,i  μ ·
1 ¨ ¸
E ª¬Ce,o,i  Pe,i , Pe, s ,i  º¼ ³c
0
e, o,i  Pe, s ,i  Pe,i  ˜
2πI
e © 2I ¹ dPe,i (119)

It is possible to use the variable change theorem for integrals as follows:

Pe,i  μ
U o Pe,i U 2I  μ (120)
2I

dPe,i
dU o dPe,i dU 2I (121)
2I
The integral limits for the variable change integral can be defined as:
0 μ
if Pe,i 0 o Ua (122)
2I

Pe, s ,i  μ
if Pe,i Pe, s ,i o U b (123)
2I
In this way, it is possible to develop (119):
Ub

³ c P  U 2I  μ  ˜
1
E ª¬Ce,o,i  Pe,i , Pe, s ,i  º¼ eU ˜ 2I dU
2
e, o ,i e, s ,i
Ua
2πI
(124)
 Pe, s ,i  μ 
Ub Ub
2I
Ua
³c e , o ,i
π
˜e U 2
dU  ³c
Ua
e , o ,i
π
˜ Ue U 2
dU
 Uncertainty cost functions for solar photovoltaic generation 195

Using the identities (69) and (138), the expression of the integral is obtained:
ce,o,i
E ª¬Ce,o,i  Pe,i , Pe, s ,i  º¼  Pe, s ,i  μ   erf U b   erf U a  
2
(125)
ce,o,iI 2 1  U a2
 eU b 
2
 ˜ e
π 2

Returning to the original limits given by the variable change [(122) and (123)], it is
possible to obtain the following expression for the expected penalty cost due to
overestimate:
ce,o,i § § Pe, s ,i  μ · § μ ··
E ¬ªCe,o,i  Pe,i , Pe, s ,i  ¼º  Pe, s ,i  μ  ¨ erf ¨ ¸  erf ¨ ¸¸
2 © © 2I ¹ © 2I ¹ ¹
(126)
§ § μ ·
2
§ Pe ,s ,i  μ ·
2
·
ce,o,iI ¨ ¨© ¸
2I ¹
¨ I¸ ¸
 ˜©e e © 2 ¹ ¹
2π
It is known that f(x)= erf(x) is an odd function, then the final expression is achieved by:
§ § μ  Pe, s ,i ·
 Pe, s ,i  μ  ¨ erf ¨§
ce,o,i μ ·
E ¬ªCe,o,i  Pe,i , Pe, s ,i  ¼º ¸  erf ¨ ¸
2 © © 2I ¹ © 2I ¹
(127)
§ § Pe ,s ,i  μ ·2 § μ ·
2
·
ce,o,iI ¨ ¨© 2I ¸¹ ¨ ¸
2I ¹ ¸
 ˜©e e © ¹
2π
In this way, it is possible to get the UCF for PEVs by adding the equations (117) and
(127).

4 Monte Carlo simulation for determinate the expected value of UCF for
PVG, WEG and PEVs

Monte Carlo simulation is a technique used to model a physical system or stochastic

process composed of random variables, which have probability density functions, as
shown in Cardell and Anderson (2010), Anderson and Cardell (2009), and Thomopoulos
(2013). The simulation is mainly used to study the behaviour of complex
nondeterministic systems, in order to generate random numbers computationally
(however, other methods may be also used), and so predict system behaviour.
In this paper, Monte Carlo simulation is used to find the uncertainty cost of
generating by incorporating photovoltaic generation, wind power generation, and electric
vehicles in a power system. The number of scenarios for the Monte Carlo simulations
was selected after test different number of iterations and contrast the analytical expected
value with the Monte Carlo expected value of the uncertainty cost. With 10^5 scenarios,
the two mentioned expected cost were very similar.
196 J.C. Arevalo et al.

4.1 Monte Carlo simulation for PVG case

The irradiance PDF data of a geographical zone is used as case base. This PDF has a
log-normal behaviour, as shown in Chang (2010). Considering the scenarios of the
generated power given by fWPV (WPV ) PDF, the costs are calculated due to underestimate
or overestimate for each scenario, in order to obtain a total cumulative cost. The main
steps used to make the simulation are set out below:
1 a power value is fixed representing the scheduled PV power set by ED model in
generator i (WPV,s,i)
2 a Monte Carlo scenario is generated through a random irradiance value generated for
generator i (Gi) according to the log-normal probabilistic behaviour
3 given the irradiance Gi, the value of power generated (WPV,i) is determined, according
to equations (9) and (10)
4 in this Monte Carlo scenario, the cost is evaluated: if WPV,s,i < WPV,i then the equation
due to underestimate (4) is used, and if WPV,i < WPV,s,i then the equation due to
overestimate (6) is used
5 the steps from 2 to 4 are repeated for certain number of Monte Carlo scenarios
6 after repeat the step 5 certain number of Monte Carlo scenarios, it is possible to
create the histogram for the total cumulative cost equivalent to the UCF
7 the expected value of the total cumulative cost is calculated; this value is the
expected value of the UCF.
To test the analytical formula for UCF versus the expected value given by the mentioned
steps, the data in Table 1 are set as inputs to the Monte Carlo simulation in the PVG case.
Table 1 Input data in PVG case

Inputs
Symbol Parameter Value
WPVr Rated active power of the PVG source [MW] 65
2
Gr Rated irradiance of the geographical environment [W/m ] 1,000
Rc Reference irradiance value [W/m2] 150
WPV,∞ Maximum power output [MW] 100
λ Location parameter of the log-normal distribution 6
E Scale parameter of the log-normal distribution 0.25
N Number of iterations 100,000
WPV,s,i Scheduled PV power [MW] 45
cPV,u,i Penalty cost coefficient due to underestimate [$/MW] 30
cPV,o,i Penalty cost coefficient due to overestimate [$/MW] 70
 Uncertainty cost functions for solar photovoltaic generation 197

The Monte Carlo simulation in PVG case provides the histograms shown in Figures 1
and 2.

Figure 1 Monte Carlo simulation for PVG case: histograms for irradiance, power, cost due to
overestimate, and cost due to overestimate (see online version for colours)

Figure 2 Monte Carlo simulation for PVG case: histogram for the penalty cost (UCF) (see online
version for colours)
198 J.C. Arevalo et al.

4.2 Monte Carlo simulation for WEG case

The wind speed PDF data of a geographical zone is used as case base. This PDF normally
has a Rayleigh behaviour, as shown in Chang (2010), Zhao et al. (2012), and Surender
et al. (2015). Considering the scenarios of the generated power given by the fW(Ww) PDF,
the cost due to underestimate or overestimate for each scenario is calculated, in order to
obtain a total cumulative cost. The main steps used to make the simulation are set out
below:
1 a power value is fixed representing the scheduled WEG power set by ED model in
generator i (Ww,s,i)
2 a Monte Carlo scenario is generated through a random wind speed value generated
for generator i (υ) according to the Rayleigh probabilistic behaviour
3 given the wind speed υ, the value of power generated (Ww,i) is determined, according
to equations (59), (61), and (63)
4 in this Monte Carlo scenario the cost is evaluated: if Ww,s,i < Ww,i then the equation
due to underestimate (4) is used, and if Ww,i < Ww,s,i then the equation due to
overestimate (6) is used
5 the steps from 2 to 4 are repeated for certain number of Monte Carlo scenarios
6 after repeat the step 5 certain number of Monte Carlo scenarios, it is possible to
obtain the histogram for the total cumulative cost equivalent to the UCF
7 the expected value of the total cumulative cost is calculated, this value is the
expected value of the UCF.
To test the analytical formula for UCF versus the expected value given by the mentioned
steps, the data of Table 2 are set as inputs to the Monte Carlo simulation in the WEG
case.
Table 2 Input data in WEG case

Inputs
Symbol Parameter Value
Ww,s,i Scheduled WEG power in generator i [MW] 100
υi Cut-in wind speed [m/s] 5
υr Rated wind speed [m/s] 15
υo Cut-out wind speed [m/s] 45
Wr Rated power output [MW] 150
ρ Lineal constant [MW/m/s] 15
κ Independent constant [MW] -75
σ Rayleigh scale parameter [m/s] 15.95
cw,u,i Penalty cost coefficient due to underestimate [$/MW] 30
cw,o,i Penalty cost coefficient due to overestimate [$/MW] 70
N Number of iterations 1,000,000
 Uncertainty cost functions for solar photovoltaic generation 199

The Monte Carlo simulation in WEG case gives the histograms shown in Figures 3 and 4.

Figure 3 Monte Carlo simulation for WEG case: histograms for wind speed, power, cost due to
underestimate, and cost due to overestimate (see online version for colours)

Figure 4 Monte Carlo simulation for WEG case: histogram for the penalty cost (UCF)
(see online version for colours)
200 J.C. Arevalo et al.

4.3 Monte Carlo simulation for PEV

The PDF of the availability power to be injected or consumed by the PEV is used as case
base. This PDF normally has a normal distribution behave according to the simulations
presented in Zhao et al. (2012). Considering the scenarios of the generated power given
by the f Pe ( Pe ) PDF, it is calculated the costs due to underestimate or overestimate for
each scenario, in order to obtain a total cumulative cost. The main steps used to make the
simulation are set out below:
1 a power value is fixed representing the scheduled PEV power set by ED model in
node i (Pe,s,i)
2 a Monte Carlo scenario is generated through a random power (Pe,i), according to the
normal probabilistic behaviour
3 in this Monte Carlo scenario the cost is evaluated: if Pe,s,i < Pe,i then the equation due
to underestimate (4) is used, and if Pe,i < Pe,s,i then the equation due to overestimate
(6) is used
4 the steps from 2 to 4 are repeated for certain number of Monte Carlo scenarios
5 after repeat the step 4 certain number of Monte Carlo scenarios, it is possible to get
the histogram for the total cumulative cost equivalent to the UCF
6 the expected value of the total cumulative cost is calculated, this value is the
expected value of the UCF.
To test the analytical formula for UCF versus the expected value given by the mentioned
steps, the data in Table 3 are set as inputs to the Monte Carlo simulation in the PEV case.
Table 3 Input data in PEV case

Inputs
Symbol Parameter Value
Pe,s,i Scheduled power to node i of PEVs [MW] 19
µ Mean [MW] 19.54
I Standard deviation [MW] 0.54
ce,u,i Penalty cost coefficient due to underestimate [$/MW] 30
ce,o,i Penalty cost coefficient due to overestimate [$/MW] 70
N Number of iterations 1,000,000

The Monte Carlo simulation in PEV case provides the histograms shown in Figure 5 and
Figure 6.
 Uncertainty cost functions for solar photovoltaic generation 201

Figure 5 Monte Carlo simulation for PEV case: histograms for power, cost due to overestimate,
and cost due to overestimate (see online version for colours)

Figure 6 Monte Carlo simulation for PVG case: histogram for the penalty cost (UCF) (see online
version for colours)
202 J.C. Arevalo et al.

5 Analysis of results and discussion

This section expects to show the results obtained analytically through the UCF calculated
by equations of Section 3, and the results of the UCF expected cost by the Monte Carlo
method (Section 4), for specific cases of PVG, WEG, and PEV.

5.1 PVG results

The result of the expected penalty cost obtained in the Monte Carlo simulation is called
MCPV (Monte Carlo cost); the result of evaluating UCF through the analytical equations
is called ACPV (analytical cost).

5.1.1 Monte Carlo cost for PVG case

The expected value is basically the average of a set of measurements (Mendenhall et al.,
2006). Therefore, when taking the average of the simulated data, the expected value of
the uncertainty cost is obtained. From the data of Figure 2, the average is obtained:
MCPV $1.2629e+03

Figure 7 Penalty cost (UCF) vs scheduled PV power (see online version for colours)

5.1.2 Analytical cost for PVG case

The total analytical penalty cost corresponding to the sum of evaluating the costs due to
underestimate and overestimate in the PVG case is showed, that is to say, replacing the
parametric variables given by the specific values of Table 1.
ACPV under estimated PV  over estimated PV (128)
 Uncertainty cost functions for solar photovoltaic generation 203

ACPV E ¬ªCPV ,u ,i WPV , s ,i , WPV ,i  ¼º  E ¬ªCPV ,o,i WPV , s ,i , WPV ,i  ¼º (129)

ACPV $1.2628e+03 (130)

The cost found by Monte Carlo has an error of 0.0072%, which shows the accuracy of
analytical calculation. In Figure 7 is presented the shape of the UCF under different
scheduled PV powers.

5.2 WEG results

The result of the expected penalty cost obtained in the Monte Carlo simulation is called
MCw (Monte Carlo cost); the result of evaluating UCF through the analytical equations is
called ACw (analytical cost).

5.2.1 Monte Carlo cost for WEG case

From the data of Figure 4 an expected value is obtained:
MCw $2.0843e+03

Figure 8 Penalty cost (UCF) vs scheduled WEG power (see online version for colours)

5.2.2 Analytical cost for WEG case

The total analytical penalty cost corresponding to the sum of evaluating the costs due to
underestimate and overestimate in the WEG case is showed, that is to say, replacing the
parametric variables given by the specific values of the Table 2.
ACw over estimated w  under estimated w (131)
204 J.C. Arevalo et al.

ACw E ¬ªCw,o,i Ww, s ,i , Wi  ¼º  E ¬ªCw,u ,i Ww, s ,i , Ww,i  ¼º (132)

ACw $2,0850e+03 (133)

The cost found by Monte Carlo has an error of 0.0355%, which shows the accuracy of
analytical calculation. In Figure 8 is presented the shape of the UCF under different
scheduled WEG powers.

5.3 PEV results

The result of the expected penalty cost obtained in the Monte Carlo simulation is called
MCe (Monte Carlo cost), the result of evaluating UCF using the analytical equations is
called ACe (analytical cost).

5.3.1 Monte Carlo cost for PEV case

From the data of Figure 6, the expected value is obtained:
MCe $20.6992

Figure 9 Penalty cost (UCF) vs scheduled PEV power (see online version for colours)

5.3.2 Analytical cost for PEV case

The total analytical penalty cost corresponding to the sum of evaluating the costs due to
underestimate and overestimate in the PEV case is showed, that is to say, replacing the
parametric variables given by the specific values of Table 3.
 Uncertainty cost functions for solar photovoltaic generation 205

ACe over estimated e  under estimated e (134)

ACe E ¬ªCe,o,i  Pe,o,i , Pe, s ,i  ¼º  E ¬ªCe,u ,i  Pe,i , Pe, s ,i  ¼º (135)

ACe $20.6990 (136)

The cost found by Monte Carlo has an error of 0.0009%, which shows the accuracy of
analytical calculation. In Figure 9 is presented the shape of the UCF under different
scheduled PEV powers.

6 Conclusions

Given the uncertainty caused by the incorporation of PVG, WEG, and PEV, a power
system operator must comprise tools or methodologies that minimise the risk associated
with the scheduling of these variable sources or loads in the ED of electricity. The
analytical developments presented in this paper may be an important part of the
mentioned tools and methodologies. The mathematical formulation presented can be
incorporated in the power system optimisation techniques in order to obtain stochastic
economic dispatch models.
Power systems are developed in an economic environment where power is scheduled.
It is necessary that the scheduled power quantities be supplied reliably; otherwise,
penalty costs are generated due to underestimate or overestimate the power available
(by renewable sources). In this paper, theoretical aspects of energy systems with PVG,
WEG and PEV are considered in order to model the stochastic behaviour of the
mentioned energy agents. With this consideration, it is possible to obtain mathematical
expressions of the penalty costs and get UCFs. UCF allow us to assess the
underestimated or overestimated costs of generated power and thus estimate the sales
price of energy; therefore, it can assess the economic viability of the generators based on
renewable energy and ensure stability in the electricity market.
From the point of view of the system operator, costs for underestimate power are
paid to the generating agent in order to compensate for losses, in case that the generator
has more power available than that scheduled in the economic dispatch. Costs for
overestimate are paid to a new scheduled generating agent, due to there were not enough
power to supply the scheduled power in the original generating agent. The analytical
equations of UCF set forth herein formulation and Monte Carlo simulation considered the
underestimate and overestimate conditions and were developed parametrically. They can
be applied to different cases of PVG, WEG and PEV, varying input parameters such as
the maximum power supplied by the generator, average irradiance value of a specific
field, the underestimate cost factor, among others. The proposed development is therefore
flexible for possible future implementations.
The results can be considered accurate, based on the comparison between analytical
results and results of the Monte Carlo simulation; the maximum error in the cases
analysed in this paper was of 0.0355%. Thus, an accurate estimate of the costs of
uncertainty, due to the stochastic behaviour of primary energy sources, allows optimising
the use of resources of power generation.
206 J.C. Arevalo et al.

References
Anderson, C.L. and Cardell, J.B. (2009) ‘Analysis of wind penetration and network reliability
through Monte Carlo simulation’, The 2009 Winter Simulation Conference (WSC), Austin,
TX, doi:10.1109/WSC.2009.5429298.
Arango, D., Urrego, R. and Rivera, S. (2017) ‘Robust loss coefficients: application to power
systems with solar and wind energy’, Int. J. of Power and Energy Conversion, Vol. 8, No. 4,
pp.221–252.
Cardell, J.B. and Anderson, C.L. (2010) ‘Analysis of the system costs of wind variability through
Monte Carlo simulation’, The 43rd Hawaii International Conference on System Sciences
(HICSS), Honolulu, HI, doi:10.1109/HICSS.2010.61.
Chang, T. (2010) ‘Investigation on frequency distribution of global radiation using different
probability density functions’, Int. J. Appl. Sci. Eng, Vol. 8, No. 2, pp.99–107.
Guo, Q., Han, J., Yoon, M. and Jang, G (2012) ‘A study of economic dispatch with emission
constraint in smart grid including wind turbines and electric vehicles’, The Vehicle Power and
Propulsion Conference (VPPC), 9–12 October 2012, doi:10.1109/VPPC.2012.6422753.
Kamankesh, H., Agelidis, V.G. and Kavousi-Fard, A. (2016) ‘Optimal scheduling of renewable
micro-grids considering plug-in hybrid electric vehicle charging demand’, Energy, Vol. 100,
No. 1, pp.285–297.
Kantar, Y.M.., Usta, I., Arik, I. and Yenilmez, I. (2016) ‘Distributions of wind speed at different
heights’, 2016 International Conference on Engineering and MIS (ICEMIS), Agadir,
doi: 10.1109/ICEMIS.2016.7745320.
Mendenhall, W., Wackerly, D. and Scheaffer, R.L. (2006) Estadística Matemática con
aplicaciones, 4th ed., ISBN: 970-625-016-6, Iberoamericana, USA.
Meng, J., Li, G. and Du, Y. (2013) ‘Economic dispatch for power systems with wind and solar
energy integration considering reserve risk’, Power and Energy Engineering Conference
(APPEEC), 8–11 December 2013, doi:10.1109/APPEEC.2013.6837207.
Mojica-Nava, E., Rivera, S. and Quijano, N. (2017) ‘Game-theoretic dispatch control in microgrids
considering network losses and renewable distributed energy resources integration’,
IET Generation, Transmission and Distribution, Vol. 11, No. 13.
Ruiz-Arias, J.A., Gueymard, C.A., Quesada-Ruiz, S., Santos-Alamillos, F.J. and Pozo-Vázquez, D.
(2106) ‘Bias induced by the AOD representation time scale in long-term solar radiation
calculations. Part 1: sensitivity of the AOD distribution to the representation time scale’, Solar
Energy, Vol. 137, No. 1, pp.608–620.
Sufen, T., Youbing, Z. and Jun, Q. (2012) ‘Impact of electric vehicles as interruptible load on
economic dispatch incorporating wind power’, The International Conference on Sustainable
Power Generation and Supply (SUPERGEN 2012), 8–9 September 2012, doi:10.1049/
cp.2012.1816.
Surender, S., Bijwe, P.R. and Abhyankar, A.R. (2015) ‘Real-time economic dispatch considering
renewable power generation variability and uncertainty over scheduling period’, IEEE Systems
Journal, Vol. 9, No. 4, pp.1440–1451.
Thomopoulos, N. (2013) Essentials of Monte Carlo Simulations. Statistical Methods for Building
Simulation Models, ISBN 978-1-4614-6021-3, Springer Science, Illinois Institute of
Technology, New York.
Xie, F., Huang, M., Zhang, W. and Li, J. (2011) ‘Research on electric vehicle charging station load
forecasting’, International Conference on the Advanced Power System Automation and
Protection (APAP), doi:10.1109/APAP.2011.6180772.
Yao, W., Chung, C.Y., Wen, F., Qin, M. and Xue, Y. (2016) ‘Scenario-based comprehensive
expansion planning for distribution systems considering integration of plug-in electric
vehicles’, IEEE Transactions on Power Systems, Vol. 31, No. 1, pp.317–328.
 Uncertainty cost functions for solar photovoltaic generation 207

Zhang, N., Behera, P.K. and Williams, C. (2013) ‘Solar radiation prediction based on particle
swarm optimization and evolutionary algorithm using recurrent neural networks’, The IEEE
International Systems Conference (SysCon), 15–18 April 2013, doi:10.1109/
SysCon.2013.6549894.
Zhao, J.H., Wen, F., Dong, Z.Y., Xue, Y. and Wong, K.P. (2012) ‘Optimal dispatch of electric
vehicles and wind power using enhanced particle swarm optimization’, IEEE Transactions on
Industrial Informatics, Vol. 8, No. 4, pp.889–899.

Appendix A

Variable change theorem

Supposes fx(x) the probability density function of x, and y = g(x), then the probability
density function of y is:
dg 1 ( y )
f y ( y) f x  g 1 ( y )  (137)
dy

Appendix B

Error function identities

b
π
³e
a
t 2
˜ dt
2
 erf (b)  erf (a)  (138)

b
π
³e  erf (b  k )  erf (a  k ) 
2
t  k 
˜ dt (139)
2
a

View publication stats