POLITECNICO DI TORINO
Repository ISTITUZIONALE
A Distributed IoT Infrastructure to Test and Deploy Real-Time Demand Response in Smart Grids
Original
A Distributed IoT Infrastructure to Test and Deploy Real-Time Demand Response in Smart Grids / Barbierato, Luca;
Estebsari, Abouzar; Pons, Enrico; Pau, Marco; Salassa, FABIO GUIDO MARIO; Ghirardi, Marco; Patti, Edoardo. - In:
IEEE INTERNET OF THINGS JOURNAL. - ISSN 2327-4662. - (In corso di stampa).
Availability:
This version is available at: 11583/2712099 since: 2018-08-30T16:28:38Z
Publisher:
IEEE
Published
DOI:10.1109/JIOT.2018.2867511
Terms of use:
openAccess
This article is made available under terms and conditions as specified in the corresponding bibliographic description in
the repository
Publisher copyright
ieee
copyright 20xx IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other
uses, in any current or future media, including reprinting/republishing this material for advertising or promotional
purposes, creating .
(Article begins on next page)
11 October 2018
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
1
A Distributed IoT Infrastructure to Test and Deploy
Real-Time Demand Response in Smart Grids
Luca Barbierato, Abouzar Estebsari, Enrico Pons, Marco Pau, Fabio Salassa, Marco Ghirardi and Edoardo Patti
Abstract—In this paper, we present a novel distributed
framework for real time management and co-simulation of
Demand Response (DR) in smart grids. Our solution provides a (near-) real-time co-simulation platform to validate new
DR-policies exploiting Internet-of-Things approach performing
software-in-the-loop. Hence, the behavior of real-world power
systems can be emulated in a very realistic way and different
DR-policies can be easily deployed and/or replaced in a plugand-play fashion, without affecting the rest of the framework.
In addition, our solution integrates real internet-connected smart
devices deployed at customer premises and along the Smart Grid
to retrieve energy information and send actuation commands.
Thus, the framework is also ready to manage DR in a real-world
Smart Grid. This is demonstrated on a realistic smart grid with
a test case DR-policy.
Index Terms—Internet-of-Things, Smart Grid, Demand Response, co-simulation, real-time simulation, distributed infrastructure, Smart Metering Architecture
I. I NTRODUCTION
Current European electrical grids are experiencing several
issues due to the rise of electricity demand, that is growing
continuously with a predicted remarkable rate of 1.1% per
year [1]. Renewable Energy Sources (RES) integration and
diffusion increase the uncertainty in electricity dispatch plan
and control due to the intrinsic fluctuation in the electricity
generation pattern of these resources. To cope with these
issues, different solutions have been proposed that ensure
reliability of the grid management, such as physical storage [2]
(e.g. batteries) and virtual power plants [3].
Even if physical storage is assessed to be of great value
for grid management, the cost of large storage systems is
too high for a massive deployment in electrical grids. On
the other hand, virtual power plants are considered a costeffective and feasible way to solve issues on grid management.
Demand Response (DR) [4] permits achieving a temporary
virtual power plant by changing the energy consumption
pattern of consumers to fulfil grid operation requirements or
economical incentives. In literature, DR policies are classified
depending on possible objectives [5], [6] (e.g. economy driven,
L. Barbierato and E. Patti are with the Department of Control and Computer
Engineering, Politecnico di Torino, Torino, Italy.
A. Estebsari and E. Pons are with the Department of Energy, Politecnico
di Torino, Torino, Italy.
F. Salassa and M. Ghirardi are with the Department of Management and
Production Engineering, Politecnico di Torino, Torino, Italy.
M. Pau is with the Institute for Automation of Complex Power Systems,
RWTH Aachen University, Aachen, Germany
Emails: {luca.barbierato, abouzar.estebsari, enrico.pons, fabio.salassa,
marco.ghirardi, edoardo.patti}@polito.it, {mpau}@eonerc.rwth-aachen.de
This work was supported by FLEXMETER, which is an EU Horizon 2020
project under grant agreement no. 646568.
reliability driven and ancillary services driven) or based on
control strategy [7], [8], [9] (e.g. based on prices, demanddispatch, direct and autonomous control). End-user flexibility
is often exploited through day ahead or intra-day scheduling to
balance the network, but (near-) real-time DR programs may
also exist to deal with unexpected contingencies or loss of
reliability in the network. Most of the current DR plans refer to
large commercial and industrial customers [10] that sell their
electrical flexibility in energy markets. However, residential
DR is becoming an important key point in grid management,
as electricity demand of households will reach 23% of the total
electricity demand in the next decade [1]. DR programs also
involving residential customers are already active in US [11],
while in Europe they are not yet deployed, mainly due to the
lack of a suitable regulatory framework. However, it is well
recognized that DR can be a strategic tool to improve energy
efficiency and to have optimal use of the grid assets [12], [13].
DR can be thus expected to gain importance also in Europe
in the next future and to play a key role for addressing the
challenges brought by the evolving scenario of the distribution
grids.
To provide a better management of distribution networks
and RES, future Smart Grids will be equipped with pervasive
Internet connected devices following the rising Internet-ofThings (IoT) approach. In the IoT vision, anything can be
connected anywhere into the global information network (i.e.
the Internet) at anytime [14]. Thus, IoT will be a preferred
medium to transmit sensed information and actuation commands [15] enabling a two-way communication among Smart
Grids devices and components [16], [17], [18]. This combination of Smart Grid view together with distributed RES and
IoT technologies is also known as Internet-of-Energy [19], [8].
These infrastructures have to be designed to deal with Big
Data transmission and processing [15]. Furthermore, software
components of these infrastructures can be optionally deployed
in cloud systems [20]. Cloud computing offers different solutions to access shared pools of configurable ICT resources
(e.g. servers, storages and applications), which can be easily
and quickly provisioned with minimal management effort.
In this scenario, Advanced Metering Infrastructure
(AMI) [21] and IoT devices (i.e. smart meters [22] and smart
appliances), are key technologies to foster novel services in
Smart Grid [23], such as the DR in residential contexts. On
these regards, smart devices are part of AMI and allow a
fine grained collection of energy measurements. In particular,
novel smart meters [22] can sample data spanning in the range
of 1s to 15min [24], depending on the services to provide.
Disaggregating and post-processing these measurements
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
Energy Aggregation Platform
Advanced Multi Metering Infrastructure
Internet
Real-Time
Simulator
Smart Devices
2
be fed with real energy information coming from real IoT
devices installed across the distribution network providing a
real-time management of DR in smart grids. The rest of this
paper is organized as follows. Section II reviews most relevant
literature solutions. Section III describes in-depth each module
of our proposed framework. Section IV introduces the DR
algorithm that we used as a test-case. Section V presents
both case study and experimental results. Finally, Section VI
discusses concluding remarks.
Fig. 1. Scheme of the proposed framework
II. R ELATED W ORK
allow retrieving information about consumption behavior
in households, such as appliance activation and energy
usage [25]. These post-processed data can feed other novel
energy services, such as DR in households [26].
In this paper, we present a novel framework (see Fig. 1)
which can serve DR and Demand Side Management (DSM)
functions in smart grids from two aspects: i) providing a
(near-) real-time co-simulation platform for validation of new
algorithms through a so-called real-time software-in-the-loop
set-up which emulates the behavior of the real-world system
very realistically; ii) integrating real IoT devices deployed at
customer premises to retrieve and collect energy information.
This framework integrates our proposed cloud-based Advanced
Multi-metering Infrastructure (a.k.a. F LEXMETER) with one
or more Digital Real-Time Simulators (DRTS) which accommodate grid models. Considering the capability of parallel
computation in Real-Time Simulation (RTS) through clustering several DRTS, and according to the new experiments on
geographically distributed real time co-simulations ([27], [28],
[29]) in which distant laboratories interconnect their DRTS
to enhance calculation power of the integrated simulation
platform, large electricity grids with a lot of elements can
be modeled and simulated for ex-ante laboratory tests. This
capability allows replacing DRTSs with the real-world Smart
Grid environment where the proposed framework is intended
to be used too. It provides a bidirectional (near-) real-time
communication with real IoT devices, and enables the interoperability among heterogeneous technologies. Regarding
the simulation aspect of our framework, the novelty can be
summarized as: i) different DR algorithms can be easily
assessed in a very realistic testbed in a plug-and-play fashion;
ii) interoperability of DR algorithms with other smart grid
control and management strategies can be evaluated without integrating all modules in a single monolithic software
or program; iii) integrating DRTS provides a very accurate
and efficient simulation of smart grid; iv) different modules,
hosting control algorithms or time-variant parameters of smart
grids, connected to this framework can be integrated while running on different Software/Hardware platforms; v) IoT based
approach used in the framework architecture enables clustering
several DRTSs to enhance overall computation power and
provide concurrent simulation of different levels/subsystems of
a smart grid, e.g. Medium Voltage (MV) grids, Low Voltage
(LV) grids and residential households. Moreover, exploiting
F LEXMETER infrastructure, the simulated scenario in RTS can
New generation smart meters are key enabler of AMI and
foster new energy related services such as DR. In [30], [31],
authors present two Smart Meter Systems that allow a bidirectional communication with a centralized DR management
platform. Aguirre et al. [30] present a new generation smart
meter designed to support new requirements for operation and
control of the distribution network grid. LeMay et al. [31]
describe a Meter Gateway Architecture for integrated control
of loads by energy aggregators. Not only Smart Meters are
important in such context. For instance, Mashima et al. [32]
present a DR System Framework leveraging on a DR client
mobile app able to directly control IoT devices according to
user policies.
Different DR Infrastructures have been proposed to cope
with the necessity of solutions for Smart Grid management
and control. In [33], Kim et al. introduce a cloud-based DR
platform able to perform power reduction requests and to
return the price incentive per customer to obtain it. Jacobsen
et al. [34] present SEMIAH, a scalable infrastructure for residential DR exploiting shiftability of smart appliances within
a Home Energy Management Gateway in communication
with a distributed platform that permits receiving commands
from a DR system to postpone appliance activation. In [35],
Bhattarai et al. designed a hierarchical DR architecture. This
infrastructure is able to perform three levels of control and
actuation to manage residential DR resources (e.g. Plug-in
Hybrid Electric Vehicle) using in each level a different kind of
DR techniques (i.e. Direct Load Control, Price based, Demand
Dispatch and Autonomous).
Müller et al. [36] highlighted the important role of the
Information and Communication Technology (ICT) in simulating future power systems. Indeed, Smart Grids are complex
systems where different entities cooperate by exchanging
heterogeneous information. In this view, all the previous
solutions need to be assessed in a simulated environment to
confirm DR service feasibility in terms of network communication, data management and result on Smart Grid beahviour.
Yang et al. [37] propose a co-simulation environment to
validate distributed controls in Smart Grid. To build closedloop models, they develop controllers in MATLAB/Simulink
that communicate with power plants models through either
UDP or TCP protocols. Manbachi et al. [38] present a cosimulation platform for evaluating the performance of a voltVAR optimization engine. In its core, it integrates a realtime simulator for distribution networks with physical devices
for Measurement & Control. MOSAIK [39], [40], [41] is
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
3
TABLE I
C OMPARISON AMONG OUR CO - SIMULATION INFRASTRUCTURE AND
LITERATURE SOLUTIONS
Solution
Our framework
Aguirre et al. [30]
LeMay et al. [31]
Mashima et al. [32]
Kim et al. [33]
Jacobsen et al. [34]
Bhattarai et al. [35]
MOSAIK [39], [40], [41]
GRIDspice [42]
ENEL [43]
Faria et al. [44]
Tan et al. [45]
Wijaya et al. [46]
Environment for DR
real-world
DER
Smart Grid
X
X
X
X
simulation
X
X
X
X
X
X
X
X
X
X
X
AMI Integration
Data
Actuation
collection
X
X
X
X
RTS
IoT
X
X
X
X
X
X
X
X
X
X
X
X
X
X
a flexible architecture to manage control strategies, scenario
specifications and simulation models by exploiting semantic
knowledge. To analyse power-flows in smart grids, MOSAIK
integrates a co-simulation between the software PowerFactory
and their model (in Matlab) for both Photovoltaic and Load
generations. GRIDspice [42] is a distributed platform that
combines existing simulation software for power generation
and distribution (i.e. MATPOWER and Gridlab-D). It exploits
a cloud-based architecture that allows the parallelization of
large simulations. Both MOSAIK and GRIDspice offer a flexible and scalable simulation framework but they do not provide
real-time features for short-transient phenomena. Finally, in
the context of smart grids real-time simulations, ENEL (the
biggest Italian utility) is running the POI P3 Smart Grid
project [43] to test and validate future smart grids. In particular, ENEL is testing new techniques for voltage regulation
through SCADA and Distribution Management System. Faria
et al. [44] present DemSi, a novel DR simulator allowing
studying different DR actions. Tan et al. [45] introduce Smart
Grid Common Open Research Emulator (SCORE). SCORE
permits emulating power and communication network in the
Smart Grid context to bridge the gap between testbed and
simulation. In addition, both Faria et al. [44] and Tan et al. [45]
include in their solutions Distributed Energy Resources (DER).
Another solution centered on DR assessment is DRSim [46],
a cyber-physical simulator for DR systems. DRSim is able to
model the emergent behavior of a community based on real
data traces, such as household consumption patterns, appliance
usage behaviors, IoT sensor data (e.g. motion sensor) and price
signals.
These presented solutions are not capable to perform simulation of DR policies integrating also RTS in their frameworks.
They also lack in taking advantages from the integration of
real data coming from real IoT devices (i.e. smart meters and
smart appliances) and/or AMI. Indeed, they should integrate
IoT communication paradigms and protocols to perform both
Hardware- and Software-in-the-Loop co-simulations to be
fully compliant with the Internet-of-Energy view.
Considering three DR related trends in the context of smart
grids, namely smart meter utilization for DR, deployment
of new DR and DSM infrastructures, and the necessity of
validation of new DR algorithm through reliable simulations,
we propose a generic solution to address all these trends:
our proposed real-time co-simulation framework is able to
simulate different algorithms through the Energy Aggregation Platform (EAP) that is a ”virtual box” where different
DR-policies are executed. EAP eases the replacement of a
DR-policy with another without affecting the rest of the platform. This is a novel simulation approach whose importance
is stressed in case DR developer has no interest or expertise in
grid modelling or other involved smart grid actors. The other
highlight of this co-simulation platform, thanks to its IoTbased approach, is its flexibility in integrating many different
simulation modules and modelling tools, even several real-time
simulators as a cluster of grid simulators. The latter feature has
a great added value in simulating a large multi-level system
(e.g. simulation of a system formed by an MV grid, several
LV grids, residential building units, individual households,
commercial and industrial consumers, etc.). Integrating DRTS
brings such capabilities of distributed co-simulations, either
local or geographically distant, (e.g. [28], [29]), and also of
remote testing of algorithms (e.g. [27]). This framework is
also ready to be used in a real-world Smart Grid environment
either by replacing RTS with real grid (i.e. hardware and/or
software components), or replacing virtual measurements in
the model with real data from smart meters. Indeed, the
proposed framework integrates also F LEXMETER, which is our
Advanced Metering Infrastructure. F LEXMETER already integrates different technologies and protocols used by novel commercial smart meters and smart appliances. The integration
of DRTS actually allows simultaneous simulations of realistic
Smart Grid scenarios in parallel with real data acquisition from
smart meters (load energy consumption or generation energy
production from renewable systems) to assess novel energy
services and evaluate new control strategies with much higher
accuracy than any other pure laboratory/simulation tests.
To highlight our contribution, Table I reports a comparison
of our framework with reviewed literature solutions for DR
co-simulation. It highlights: i) if the environment to run DRpolicies consists on simulation tools or they can be applied
also in a real world smart grid or they integrate DER; ii) if
AMI is integrated and provides features to collect data from
smart devices and send actuation commands; iii) if a Real-time
simulator is used; and iv) if the infrastructure interacts with
IoT devices (i.e. smart meters and smart appliances).
III. D ISTRIBUTED F RAMEWORK FOR D EMAND R ESPONSE
In this section, we describe the components of the proposed
distributed framework. As shown in Fig. 1, we identify three
main components: i) Advanced Multi-metering Infrastructure,
ii) Energy Aggregation Platform and iii) Real-Time Simulator.
A. Advanced Multi-metering Infrastructure
The Energy Aggregation Platform leverages upon our Advanced Multi-metering Infrastructure, called F LEXMETER,
that fosters general purpose services in a smart grid scenario.
We presented F LEXMETER in our previous work [47], where
we described a solution for distributed state estimations that
works together with a service for automatic network reconfiguration. Whilst, in this paper we describe a novel solution for (near-) real-time management and co-simulation of
Application
Layer
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
Non-Intrusive
Load Monitoring
User Awareness
4
Energy Aggregation
Platform
Middleware
Layer
REST API Interface Manager
Device Manager
Assets Manager
Communication Engine
Inbound Pipeline
Outbound Pipeline
Event Sources
Command Destinations
Data storage
Technology
Integration
Layer
Message Broker
Hardware Data Sources
Real-Time Simulator
OPAL RT
Wi-Fi
ZigBee
PLC
6LowPan
RTDS
Fig. 2. Scheme of F LEXMETER infrastructure
DR-policies in smart grids. Thus, Demand Response through
Energy Aggregation Platform is another service leveraging
upon F LEXMETER that is briefly introduced in this section
to give a complete overview of the proposed solution for
co-simulation of DR algorithms.
F LEXMETER’s architecture has been designed to cope with
interoperability among heterogeneous smart devices abstracting different underlying low-level technologies. This enables
a fine-grain monitoring of the overall energy distribution
network, offering a common data access to this large amount
of incoming information. This provides Energy Aggregation
Platform with necessary information (i.e. appliance start-time
and power profile of buildings or substations) to perform DR
services. F LEXMETER has been designed as a single instance
that can optionally run in a cluster of servers or in a cloud
system allowing multi-tenancy [48].
As shown in Fig. 2, F LEXMETER is a three-layered architecture with: i) a Technology Integration Layer, ii) a Middleware
Layer and iii) an Application Layer.
1) The Technology Integration Layer is the lower layer of
F LEXMETER. It is made up of different Technology Integration
Adapters (TIA), one for each technology, developed following
a methodology proposed in [49]. TIA is a middleware-based
software component in charge of integrating heterogeneous
hardware data sources that exploit different communication
protocols (e.g. IEEE 802.11, ZigBee or 6LowPan). It abstracts device features and functionalities providing common
interfaces to access them. In addition, TIAs harmonize information coming from low-level devices in a common dataformat before sending it to the rest of the platform via
MQTT [50]. MQTT stands for Message Queuing Telemetry
Transport. It is a communication protocol that implements the
publish/subscribe paradigm [51]. We developed also specific
TIAs to integrate DRTSs, such as RTDS and Opal-RT. Virtual
devices simulated on DRTS are transparently seen by the other
F LEXMETER’s actors as IoT devices ready to communicate
over the Internet. Thus for the infrastructure, there is no
difference in dealing with a real or a virtual IoT devices. This
allows i) an easy switch from a simulation environment to a
real-world smart grid, and vice-versa, and ii) a bidirectional
data exchange between real-world devices and RTS, even in
(near-) real-time.
2) The Middleware Layer in Fig. 2 consists of several
software components acting together to: i) ensure bidirectional
communication with TIAs; ii) collect, check and store data
messages; iii) provide unified interfaces to access data, devices, assets and maintenance operations and iv) send commands to devices. The Message Broker manages the MQTT
communication with devices allowing asynchronous bidirectional communication with the cloud infrastructure. It routes
all messages to the Communication Engine that manages
the communication between the platform and the devices.
In its core, Event Sources receives measurements and alerts
from TIAs, thus from devices, and checks their integrity
before pushing them into the Inbound Pipeline that prevents
congestions in storing information into the Data Storage.
The Outbound Pipeline manages out-coming messages to be
sent to devices through the Command Destinations. The Data
Storage is a module that integrates different non-relational
database models to store information. The Device Management
supervises the device provisioning handled by the application
layer. Whilst, the Asset Modules handle information of the
assets in F LEXMETER (e.g. people, places and things). Finally,
applications retrieve data and manage entities or devices
from the platform through the REST API Interface Manager
that provides REST (Representational State Transfer) web
services [52].
3) The Application Layer offers APIs and tools to develop
services and applications to manage and post-process information coming from the underlying layers. For example, User
Awareness service allows end-users in retrieving information
about their energy behaviour patterns [53]. Non-Intrusive Load
Monitoring[25] disaggregates home appliances load consumption from a single point of measurement given by the smart
meter. Energy Aggregation Platform is a complex service to
perform DR in Smart Grid (see Section III-B). This layer
promotes also the data exchange among services.
B. Energy Aggregation Platform
The Energy Aggregation Platform (EAP) is a software
designed to provide novel services for smart grid management
(i.e. DR and DSM) to different stakeholders playing in the energy marketplace (e.g. Distribution System Operators, Retailers and Energy Aggregators). EAP is a ”virtual box” that gives
to the whole framework the flexibility to deploy or replace
easily a DR-policy without affecting the rest of the framework.
EAP exploits the F LEXMETER infrastructure that provides
APIs and tools to abstract functionality of devices, either real
or virtual, and transparently manage information, as pointed
out in Section III-A. In addition, EAP provides common
software interfaces to establish a bidirectional communication
among DR-policies and smart devices. Thus, DR-policies easily retrieve energy information and send actuation commands
to smart devices in (near-) real-time neglecting their low-level
technologies (i.e. hardware and protocols). EAP is developed
following the multi-tenant design pattern approach [48]. As
shown in Fig. 3, it consists of five modules described in the
following.
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
Multi-Tenant Proxy
Energy API Manager
Algorithm Manager
EA Data
Storage
Demand Side
Management Policy
Demand
Response Policy
REST Client
Fig. 3. Scheme of Energy Aggregation Platform
Pending Service
Request
1
0…*
1
Household Unit
Group
1
0…*
Completed Service
Request
0…*
0…*
mands (result of DR-policies) to F LEXMETER by exploiting
the REST Client module. Then, F LEXMETER will send these
commands to the right smart devices via MQTT protocol.
This process is performed in (near-) real-time. Thanks to this
module, new DR-policies can be easily deployed or replaced
without affecting the rest of the framework.
Finally, the Multi-Tenant Proxy manages stakeholders authentications before creating a new instance of EAP (represented by the dashed-line macro-block in Fig. 3). Each
EAP tenant handles its own set of Household Unit Groups
with a specific DR-algorithm. Thus, different EAP tenants are
executed at the same time and each of them manages different
stakeholders’ customers avoiding data intermingling.
1
C. Smart grid real time simulator
1
1
0…*
Household Unit
5
0…*
1
0…*
Appliance
Fig. 4. Energy Aggregation Platform Data Model
The REST Client retrieves information on assets and measurements from F LEXMETER’s REST API Interface Manager.
Energy measurements are collected in (near-) real-time by
F LEXMETER that receives them from smart devices through
MQTT protocol. The REST Client is also used i) to populate
or update information in the EA Data Storage module and
ii) to send actuation commands to smart devices, again through
F LEXMETER.
The EA Data Storage module exploits a non-relational
database (e.g. MongoDB) and stores information needed by
DR-policies following the data model in Fig. 4. This data
model describes customers, called Household Units, involved
in the DR strategy. Each instance of Household Unit is in
relation with a set of Appliances that might participate to a
DR-policy. Household Units are grouped in Households Unit
Groups depending on the relations between customers (e.g.
customers belonging to a common substation or to a common
retailer). Each Household Unit Group is in relation with i) a set
of Pending Service Request and ii) a set of Completed Service
Request. The Pending Service Request represents an action
to be performed in a Household Unit Group (i.e. DR-event)
triggered by stakeholders. Once the action is performed by
DR-policy, the Pending Service Request becomes a Completed
Service Request that includes additional information on the
results of the fulfilled event.
The Energy API module offers REST APIs to create a
Pending Service Request and to retrieve all the information
in the EA Data Storage.
The Algorithm Manager is the ”virtual box” that contains
and executes policies for both DR and DSM events. In order
to perform the DR-policies’ routines, this module defines and
provides software interfaces to retrieve and update all the
information in the EA Data Storage (e.g. Household Units
involved in a specific DR-event). Once new actions must be
performed, the Algorithm Manager posts the actuation com-
The proposed distributed framework is ready to work in
a real Smart Grid environment. Thus, it is able to retrieve
energy data from real smart devices and send them back
actuation commands that are outputs of the DR-policy running
on EAP. To evaluate novel DR-policies and their impacts on
the Smart Grid, this framework offers a simulation engine
consisting of one or several DRTS, such as RTDS or OPALRT. In this scenario, F LEXMETER infrastructure provides
common interfaces to enable a bidirectional communication
with smart devices and collects energy information. RTS
realistically reproduces the behavior of Smart Grids with high
accuracy and efficiency. This is gained due to three main
reasons: parallel and distributed simulation capability, different
multicore processors (FPGA in OPAL-RT, or NovaCor in
RTDS) with respect to normal CPUs (in PCs), and advanced
and optimal solvers to perform discrete-time simulation with
very narrow time-steps (in range of micro- to milli- seconds)
efficiently. Using our IoT-based framework a cluster of DRTS
can concurrently run being located either in the same lab or in
several laboratories even geographically distant. This not only
increases computation power which is highly demanded by
large-scale grids with many components and actors, but also
enables concurrent simulation of different hierarchical levels
of smart grids by running on several simulators (e.g. MV grids,
LV grids, building units, households, etc.). In our framework,
hard real-time simulations are performed by DRTS, where
the power system model is running in time domain, while
the rest of the modules involved in the co-simulation are
running in (near-) real-time. Since the communication delay
in the co-simulation infrastructure is often much smaller
than modules’ responses/interactions, synchronization can be
ensured if real-time simulators are running with a shifted time
stamp. Time-variant parameters (e.g. energy consumption,
generation output, control commands, etc.) can be periodically
pushed into the running real-time model to observe Smart
Grid operational behaviour. However, such interactions are not
necessarily done with the same rate of simulation sampling:
for Hardware-In-the-Loop (HIL) a higher frequency of sampling is needed, while for Distribution Management System
applications the frequency could be lower. In our framework,
we use RTS with its narrow time-steps for mainly two reasons: enabling the platform to perform HIL tests, especially
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
with embedded system controllers or smart meters, and slow
transient analysis of the electricity grid in terms of voltage
or frequency deviation by interoperability studies of demand
response algorithms and other grid control and management
strategies (e.g. Volt-VAR control). It can also replicate the
energy consumption patterns in households exploiting the real
energy information coming in (near-) real-time from smart
devices through F LEXMETER.
As mentioned in Section III-A, RTS communicates with the
rest of the framework by exploiting a specific TIA developed
filling a methodology proposed in our previous work [54]. This
TIA implements the two communication paradigms provided
by F LEXMETER: i) publish/subscribe [51] based on MQTT
protocol [50] and ii) request/response based on REST [52].
Finally, once the DR-policy performs some actions, the
EAP translates these actions into actuation commands that are
sent to the virtual smart devices that are simulated in RTS,
through the F LEXMETER infrastructure.
For the preliminary testing and validation of the DR policies, as an alternative to real smart devices, virtual smart
meters and virtual smart appliances should be used in the
RTS environment. Therefore we developed a profile generator
able to create realistic patterns of consumption for residential customers, down to the single appliances, allowing the
emulation of the real behavior of households in the smart
grid. The profile generator can create multiple days of power
consumption data, thus enabling long term simulations, and
uses real consumption data of single appliances to generate
the overall power consumption profile. Generated data provide
the time-variant set points of active and reactive power for
each household, which are used by the DRTS engine for the
simulation of the grid operating conditions.
A dedicated software module acts as controller of the
simulation, handling the proper transmission of the household
consumption set points to the RTS. The same software module
also integrates a TIA and emulates the virtual smart devices
targeted by the DR commands sent by the EAP.
Once a DR-event is triggered, the simulation controller
module receives via MQTT, through F LEXMETER, the actuation commands issued by the EAP. It then processes the
command (e.g shut down or reactivation of an appliance) accordingly modifying the consumption profile of the appliance,
and updates the set points of the associated household, which
are subsequently forwarded to RTS.
D. Deployment considerations
As pointed out in the previous sections, the proposed
framework for Demand Response follows the modern software
design patterns (e.g. multi-tenancy [48]) to build distributed
architecture that can be deployed on cloud systems.
In this view, F LEXMETER infrastructure can be deployed
in cloud systems and offered to stakeholders as Platform as
a Service (PaaS). PaaS is a category of cloud computing
services that provides all the facilities required to support the
complete application life-cycle. Thus, different stakeholders
(e.g. Distributor System Operators, Energy Aggregators and
Retailers) develop, run and manage their applications and services without the complexity of building and maintaining the
6
whole F LEXMETER infrastructure. As an alternative, PaaS can
also be delivered providing all the software to be installed in
private data-centers managed by internal IT departments [20].
This second deployment approach could minimize the data exchange over the Internet among components (either hardware
or software) and F LEXMETER infrastructure.
EAP can be deployed in cloud systems and offered to
stakeholders as Software as a Service (SaaS). SaaS is a
software delivery methodology that provides virtualized computing resources over the Internet. SaaS architecture provides a
complete software solution, including set-up and infrastructure
management. SaaS is a software distribution model where a
producer develops, operates and manages an application, making it available to its customers via the Internet. Thus, EAP can
be given with a set of predefined DR-policies that can be easily
customized, updated or replaced by stakeholders according to
their requirements and use cases. As an alternative, EAP can
also be delivered as software to be deployed in private datacenters managed by stakeholder’s IT departments [20].
It is worth noting that the different hardware and software
components can be installed in various locations geographically closed or distant, and they communicates over the Internet following the Internet-of-Energy view. Thus, F LEXME TER platform, the different EAP instances and the real-time
simulators can be equally installed and deployed in the same
data center or in different cloud systems. These deployment
choices will not affect the behaviour of the overall proposed
framework.
IV. G REEDY A LGORITHM FOR D EMAND R ESPONSE
As mentioned in Section III-B, EAP is a ”virtual box” to
easily deploy and/or replace different DR-policies. To better
characterize the functionalities of EAP, we designed and run
experiments with an effective DR procedure belonging to the
paradigm of Greedy Algorithms [55], hereinafter referred to
as GrAl. Greedy Algorithms are solution approaches relying
on local optimal choices with the hope of finding a global
optimum. Main advantages are the easy design and the limited
computational running times. We point out that GrAl is just
one of the possible algorithms that can be embedded into
EAP ”virtual box” and ran in our framework for DR. More
complex algorithms can be developed and executed following,
for instance, the methodologies presented in [5], [7], [8].
The solution approach is based on a two-stage procedure
activated whenever a DR-event for overall exceeding energy
consumption is identified. In both stages of the procedure
the problem is modeled as a Knapsack Problem with binary
variables (KP0/1) [56], and a classical greedy approach is then
applied two times to the modeled KP0/1 problem. Once a
DR-event happens, the GrAl reads all active appliances starting
times in every Household Unit considered in the specific
scenario and computes the overall energy consumption from
the DR-event time-stamp on to a desired time horizon (e.g. 15,
30 or 60 minutes). As an input, GrAl also takes a desirable
(feasible) cumulative energy consumption, thus it is able to
compute the total amount of energy that should be cut-off
via appliances shut down. Then, for each Household Unit,
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
V. E XPERIMENTAL R ESULTS
Test results presented in the following are generated through
emulation of an LV smart grid subtending 360 residential
customers in RTS environment. The considered grid is a
portion of a real distribution network in a semi-rural area in
north of Italy, and it is composed of 20 nodes out of which 15
nodes are connected to residential customers (Fig. 5). The total
number of households is approximately equally distributed
among the nodes, so that each node subtends between 20 and
28 customers. The size and the ratio of the MV/LV transformer
in the upstream secondary substation is 250 kVA and 22/0.4
kV respectively. Resistance and reactance (R, X) of lines with
cross-sections 16, 50, and 95 mmq are (1.16, 0.0817), (0.391,
0.0779), and (0.19, 0.0751) ohm/km respectively.
For this scenario, daily power consumption profiles of
the households are created using the load profile generator
mentioned in Section III-C. For each customer, a set of appliances existing within the house has been extracted taking into
Algorithm 1 GrAl
1: INPUT: ai,j : set of appliances (j-th appliance out of ni appliances of
building i = 1..N ); fi : global discomfort of switching off building i; pij :
discomfort of switching off the j-th appliance (out of ni ) of building
i; b: maximum consumption level after the switching off. TDREvent
F and
and TON : time for beginning the switch-off/switch-on. ∆OF
T
F : time available for the switch-off/switch-on; c
∆OF
:
consumption
ijt
T
of appliance ai,j at time t.
2: for i = 1..N, j = 1..ni do
F
PTDREvent +∆OF
1
T
3:
wij = OF
ci,j,t
F
t=T
∆T
DREvent
4: end for
5: A = ∅ (Initialization of the switching off candidate set)
6: Order buildings i in non increasing order of Pnifi
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:
27:
28:
29:
30:
31:
j=0
wij
currentload = 0
for i = 1..N do
Pni
wij < b then
if currentload + j=0
Pni
wij
currentload = currentload + j=0
else
Add to set A all appliances belonging to building i.
end if
end for
p
Order ai,j ∈ A in non increasing order of wij
ij
currentload = 0
for i = 1, .., size(A) do
if currentload + wij < b then
currentload = currentload + wij
xij = 1 (appliance not to be switched off)
else
xij = 0 (selected for switching off)
end if
end for
Set A = {ai,j
P : xij = 0} (appliances to be switched off).
Set wtot = (i,j)∈A wij .
Order ai,j ∈ A in non decreasing order of pi,j .
Tcurr = TDREvent
for k = 1, .., size(A) do
Let (i, j) be the indexes of appliance Ak .
Set the switch off for appliance ai,j at time Tcurr .
∆OF F
wij
32:
Tcurr = Tcurr + wT
tot
33: end for
34: Tcurr = TON
35: for k = size(A), .., 1 do
36:
Let (i, j) be the indexes of appliance Ak .
∆ON
37:
Tcurr = Tcurr + wT wij
tot
38:
Set the switch on for appliance ai,j at time Tcurr .
39: end for
40: OUTPUT: Off/On Timings set for selected appliances.
MV / LV
250 kVA
22 / 0.4 kV
Δ Y 11
2
CU 3.5 x 95 mmq;
L = 184.3m
7
9
CU 4 x 16 mmq;
L = 13m
CU 3.5 x 95mmq;
L = 70m
CU 3.5 x 50mmq;
L = 73.8m
5
CU 3.5 x 95 mmq;
L = 22.42m
Legend:
Underground cable
Overhead line
Load
11
CU 3.5 x 50mmq; CU 4 x 16 mmq;
L = 28m
L = 24.8 m
3
1
CU 4 x 16 mmq;
L = 14.13 m
CU 4 x 16 mmq;
L = 25.7m
CU 3.5 x 95 mmq;
L = 226.2m
MV Node
GrAl sums the relative consumption of the appliances so as to
have the global consumption for each house. The algorithm
then orders Household Units considering both the global
consumption and the relative costs, represented by a single
discomfort integer parameter when selecting a specific house,
and selects the ones providing the least global discomfort for
end users with a global energy save greater than the requested
cut-off. The discomfort parameter for Household Units has
been introduced to make possible not to select every time the
same set of them. This can be achieved by modifying the
value of the parameter after a DR-event, leading to a different
ordering of Household Units during GrAl runs.
In the second stage, among the selected Household Units,
the algorithm selects the minimum set of appliances ordering
them with respect to a discomfort parameter for end-users
providing the least number of appliances that guarantee that
the global cut-off is respected. Also in this case, the discomfort
parameter for end-users is introduced to weight differently the
cost of the appliances involved in the shut-down if necessary.
The reactivation phase is performed as the inverse priority
order of appliances shut down, providing the most important
appliances to be switched on first.
The detailed GrAl procedure is presented in Algorithm 1.
The main steps are the following:
• Compute the average consumption of all appliances activated during the considered event (steps 2-4).
• Solve a KP0/1 problem on buildings, to select the involved buildings (steps 5-14)
• Solve a KP0/1 problem on all the appliances of the
selected buildings, to select the appliances to be switched
off (steps 15-24)
• Give a timing of switching off (in non-decreasing discomfort order - steps 25-33), and on again (in non-increasing
discomfort order, steps 34-39). In order to obtain a
smooth power cut in time, the time passing after one
appliance switched off to the next is proportional to the
amount of energy cut. Analogously, the time passing after
an appliance is switched on to the next is proportional to
the amount of energy added.
7
4
14
CU 3.5 x 50mmq;
L = 81.5m
13
CU 4 x 95mmq;
CU 3.5 x 95mmq; L = 37.55m
L = 34.57m
6
CU 3.5 x 95mmq;
L = 41.14m
8
CU 3.5 x 50mmq;
L = 32m
10
12
CU 4 x 16 mmq;
L = 63.2 m
15
16
CU 4 x 16 mmq;
L = 39.36 m
CU 4 x 16mmq;
L = 44m
18
19
20
17
CU 3.5 x 50mmq;
L = 25.33 m
Fig. 5. Topology and line parameters of the grid.
account user-defined percentages of diffusion of the appliance.
As an example, 100% of the customers has been assumed
to own a fridge, whereas only 50% of the customers has
been considered to have a dishwasher. The overall power
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
8
Fig. 7. Total apparent power at MV/LV substation with and without DR.
Fig. 6. Example of power consumption profile generation for a household.
consumption profile for each household is obtained as the
aggregation of the single appliances’ consumption, which
include a large set of devices such as fridge, washing machine,
dishwasher, microwave oven, TV, etc. In particular, in the performed simulations, the following appliances are considered as
smart IoT devices shiftable through DR policies: fridge, standalone freezer, washing machine, tumble dryer, dishwasher
and electric water heater. Fig. 6 shows, as an example, the
consumption profile generated for one of the households,
where the contribution of the main energy-hungry devices, like
washing machine and dishwasher, can be identified.
Given this set-up, DRTS can run the smart grid emulation.
Measurement data from the households virtual smart meters
are sent to F LEXMETER every minute thanks to the TIA
embedded in the simulation controller. In addition, time-tagged
switch-on events from the shiftable smart devices are also sent
to the cloud to provide EAP with the relevant data needed
for the proper application of the DR logic. The simulation
controller also has a specific sub-module in listening mode,
which allows the collection of the actuation commands from
the EAP whenever a DR-event is triggered. In this way, the
household consumption profiles can be updated and used to
assess the effects of the DR policy under test on the smart
grid behavior.
Fig. 8. Voltage profile at node 16.
Simultaneously, voltage profiles were also slightly improved
(Fig. 8).
The small voltage drops in LV nodes might be negligible,
and the risk of transformer congestion could be resolved by
replacing it with a larger one. But considering trends in smart
grids towards integrating more distributed energy resources
(e.g. rooftop Photovoltaic panels in urban areas), total power
demand of these LV systems from upstream networks would
remain below transformers’ capacity for longer time periods of
a day. Therefore, it is wise to utilize smart devices and efficient
DR algorithms to control the events instead of physically
reinforcing costly infrastructure.
A. Results on smart grid behavior
B. Performance analysis of GrAl
To investigate the impacts of DR on the behavior of smart
grids, two simulation scenarios, with and without DR, were
executed, and the power flow at the transformer and the voltage
profiles at different nodes were monitored and captured every
1 minute for a 24-hour period.
The results of the scenario without DR show that the
apparent power flowing through transformer is reaching the
transformer capacity (80 to 96 percent) after 18:00. At the
same time, voltage profiles of some far nodes report a slight
violation of voltage (to less than 90 percent) during the peak
load event. In the other scenario we triggered DR after 18:30
to shave peak loads. Operation of the appliances selected by
the EAP was stopped for 2 hours and, therefore, reactivation
occurs after 20:30. This load reduction during the peak time
along with a consequent loss reduction in the grid resulted in
a decrease of apparent power flow at the substation (Fig. 7).
To estimate the maximum error generated by the heuristic
algorithm of the DR-policy (see Section IV) w.r.t. to the
precise percentage quantity. The test instances have been
generated in the following way. The number of Household
Units (i.e. customers) are 25, 50, 100 and 200 to represent
different scales of the problem. The error is calculated as the
ratio between the total consumption and the largest consuming
appliance. The number of appliances is constant for each
Household Unit, namely 6 different appliances. We present
results up to 200 Household Units because these are primarily
affected by error. The worst case is thus represented by the
scenario where the appliance selected to exceed the cut-off
is the most consuming one, giving the ratio as explained.
We recall that this is a theoretical worst case analysis since
appliances, before being selected, are ordered in decreasing
order w.r.t consumption, thus the real error should be smaller.
9
Household Units Cut-off Percentage
6%
5%
4%
CPU Time
8
10% Request Cutoff
20% Request Cutoff
30% Request Cutoff
40% Requested Cutoff
7
6
10% request of cut
20% request of cut
30% request of cut
40% request of cut
5
3%
2%
s
Real Cut-off Percentage
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
1%
4
3
0%
2
25
50
100
200
Household Units
1
0
Fig. 9. Cut-off percentage exceeding nominal request.
Increasing the number of appliances leads to more tight results.
Power profiles of the different appliances have been generated
so that: i) the duration is indicated in minutes and is uniformly
randomly generated in the range [80. . . 160]; ii) the energy
consumption for each minute is uniformly randomly generated
in the range [0.5. . . 1.5] kW. The cut-off requests are 10%,
20%, 30% and 40% of the global consumption generated by
the whole set of appliances.
Fig. 9 shows the relation between the number of Household
Units of a specific cut-off request and the increase in the cutoff quantity calculated by the algorithm with respect to the
nominal ”optimal” cut-off obtained by the simple percentage
calculation. Clearly the nominal cut-off can also partially
deactivate an appliance, thing which is not possible in reality
since an appliance is switched-on or off. As can be seen, if
more Household Units are considered, the algorithm tends to
perform better since there is more room for optimization than
with fewer appliances. In general, the worst case is reached
for 25 Household Units only and never exceeds 5.5%. This
is caused by the higher impact of a single appliance w.r.t.
the global consumption. Besides, notice that this is only a
rough estimation of the percentage error as the discreteness of
appliances activation/deactivation is not taken into account at
all in this estimation.
C. Scalability performance of the framework
In this section, we present results about the scalability of the
proposed co-simulation framework. We performed stress-tests
with 1k, 5k and 10k Household Units, respectively, to evaluate
the computation performance of i) the F LEXMETER infrastructure, ii) the Energy Aggregation Platform and iii) the proposed
DR heuristic running in EAP. On average, we considered five
appliances per Household Units that can be managed. We also
present results on the performance of data transmission among
the different actors in our framework.
As mentioned in Section III-D, the various components of
the proposed framework can be installed in different locations
and they communicate over the Internet. Hence, to perform
these stress-tests, a single instance of EAP was deployed in
a physical server in our data center with a CPU Intel Xeon
3.40 GHz with 8 cores and 32 GB of RAM. Whilst, the
F LEXMETER platform was running in a virtual server in a
cloud with a CPU Intel Xeon 2.20 GHz with 16 cores and
24 GB of RAM. To perform these tests, we developed a
special TIA to simulate the communication behaviour of smart
1000
2000
5000
10000
Household Units
Fig. 10. CPU time trend w.r.t. Household Units.
appliances. EAP, F LEXMETER and TIA have been deployed
in Italy, Romania and Germany, respectively. According to our
tests, increasing the number of Household Units from 1k to
10k, thus the number of smart appliances, the usage of CPU
and RAM was almost constant for both EAP and F LEXMETER.
EAP had less than 5% CPU and RAM occupancy. Whilst,
F LEXMETER occupied CPU less than 30% and RAM between
70% and 80%.
Fig. 10 benchmarks the CPU time in seconds needed by the
DR-algorithm to perform optimization for the different classes
of requested cut-offs. Results confirm that the trend of the
CPU time is basically linear with respect to the number of
Household Units. The increase in the CPU time is consistent
with the increase in the number of appliances of the problem
instance set. This can also assess the upper limits in term of
problem sizes if the utility needs to set up a maximum time
for computing a solution of the DR. Also in these tests, the
cut-off requests are 10%, 20%, 30% and 40% of the global
consumption generated by the whole set of appliances. As
shown in Fig. 10, the computational time needed by the DR
algorithm is almost the same for the different cut-off requests
and it varies from about 0.06s to about 7.5s for 1k and 10k
Household Units, respectively.
To assess the transmission performance of the proposed
infrastructure over the Internet, we performed communication
tests to estimate latency in retrieving data and sending actuation commands. The results of these tests are shown in the
box-plots of Fig. 11.
Fig. 11-(a) shows the latency values when sending measures from the smart devices to F LEXMETER infrastructure
over MQTT protocol, with a payload of about 150 byte. As
shown by the graph, data transmission is not affected by the
variation of Household Units, maintaining a median value of
about 70ms for Household Units varying from 1k to 10k.
Furthermore, the variability of the latency values is in general
very low.
Fig. 11-(b) shows the latency values of REST Web Service when retrieving information from F LEXMETER. More
specifically, EAP invokes a specific REST Web Service to
obtain information from all the Household Units involved in
the DR-event. As shown in Fig. 11-(b), the latency is generally
very low (less than about 8s).
Fig. 11-(c) reports the latency of actuation commands sent
from EAP to smart devices through the F LEXMETER infras-
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
10
Commands latency values
REST latency values
MQTT latency values
1000
700
12
900
600
800
11
500
700
ms
s
ms
10
400
300
600
500
9
400
200
8
100
7
300
200
100
1000
5000
10000
1000
5000
10000
1000
5000
Household Units
Household Units
Household Units
(a)
(b)
(c)
10000
Fig. 11. Transmission latency at different Household Units: (a) MQTT, (b) REST, (c) Commands.
tructure. This latency includes: i) the time needed by EAP to
post the actuation commands to F LEXMETER via REST and
ii) the time needed by F LEXMETER to send such commands
to the smart devices via MQTT. Each actuation command is
intended with a payload of about 300 byte. Even in this case,
the graph shows no significant variation of latency at different
Household Units (about 150ms).
It is worth noting that we performed these stress-tests
to evaluate the correct behaviour of the overall framework.
In a realistic scenario and following the de-centralization
approach of Smart Grids management, different instances of
both EAP and F LEXMETER can be deployed in cloud systems
or data centers geographically closed to minimize the delay on
data transmission. We can also suppose that different EAP and
F LEXMETER instances can manage different portion of the
Smart Grid or different Microgrids.
VI. C ONCLUSION
In this paper, we presented a novel distributed framework
for both (near-) real-time management and co-simulation of
DR-policies in Smart Grids. We also discussed how the
components of the proposed framework can be deployed in a
distributed environment compliant with the Internet-of-Energy
view. The usability of our solution was demonstrated on a
realistic smart grid, where a test case DR-policy was applied.
In this scenario, we discussed and compared the results of total
apparent power at MV/LV substation and the voltage profiles
at different nodes with and without DR. To perform these tests,
we deployed components of our framework in Italy, Romania
and Germany discussing the advantages of having a distributed
co-simulation platform and presenting its performance.
In our view, limitations and challenges to deploy the proposed solution in real-world scenarios are related to a lack
of current regulatory frameworks instead of technological
issues. Moreover, the current marketplace does not include real
incentives for customers that will help to spread DR solutions.
Another limitation consists on rising smart appliances that are
not yet widespread at customer’s home premises, but they will
be in the future years.
R EFERENCES
[1] P. Capros, A. D. Vita, N. Tasios, D. Papadopoulos, P. Siskos, E. Apostolaki, M. Zampara, L. Paroussos, K. Fragiadakis, N. Kouvaritakis,
L. Hoglund-Isaksson, W. Winiwarter, P. Purohit, H. Bottcher, S. Frank,
P. Havlik, M. Gusti, and H. P. Witzke, EU energy, transport and ghg
emissions trends to 2050. Wiley, 2013.
[2] A. Mohd, E. Ortjohann, A. Schmelter, N. Hamsic, and D. Morton, “Challenges in integrating distributed energy storage systems into future smart
grid,” in Industrial Electronics, 2008. ISIE 2008. IEEE International
Symposium on. IEEE, 2008, pp. 1627–1632.
[3] J. S. Vardakas, N. Zorba, and C. V. Verikoukis, “A survey on demand
response programs in smart grids: Pricing methods and optimization
algorithms,” IEEE Communications Surveys & Tutorials, vol. 17, no. 1,
pp. 152–178, 2015.
[4] R. Deng, Z. Yang, M.-Y. Chow, and J. Chen, “A survey on demand
response in smart grids: Mathematical models and approaches,” IEEE
Transactions on Industrial Informatics, vol. 11, no. 3, pp. 570–582,
2015.
[5] P. T. Baboli, M. P. Moghaddam, and M. Eghbal, “Present status and
future trends in enabling demand response programs,” in Power and
Energy Society General Meeting, 2011 IEEE. IEEE, 2011, pp. 1–6.
[6] T.-C. Chiu, Y.-Y. Shih, A.-C. Pang, and C.-W. Pai, “Optimized dayahead pricing with renewable energy demand-side management for smart
grids,” IEEE Internet of Things Journal, vol. 4, no. 2, pp. 374–383, 2017.
[7] B. P. Bhattarai, B. Bak-Jensen, P. Mahat, J. R. Pillai, and M. Maier,
“Hierarchical control architecture for demand response in smart grids,”
in Power and Energy Engineering Conference (APPEEC), 2013 IEEE
PES Asia-Pacific. IEEE, 2013, pp. 1–6.
[8] C.-C. Lin, D.-J. Deng, W.-Y. Liu, and L. Chen, “Peak load shifting in the
internet of energy with energy trading among end-users,” IEEE Access,
vol. 5, pp. 1967–1976, 2017.
[9] Y. Wang, S. Mao, and R. M. Nelms, “Distributed online algorithm for
optimal real-time energy distribution in the smart grid,” IEEE Internet
of Things Journal, vol. 1, no. 1, pp. 70–80, 2014.
[10] F. Y. Xu and L. L. Lai, “Novel active time-based demand response for
industrial consumers in smart grid,” IEEE Transactions on Industrial
Informatics, vol. 11, no. 6, pp. 1564–1573, 2015.
[11] US Department of Energy, “Benefits of demand response in electricity
markets and recommendations for achieving them,” US Department of
Energy, Washington, DC, USA, Tech. Rep, 2006.
[12] P. Bertoldi, P. Zancanella, and B. Boza-Kiss, “Demand response status
in eu member states,” Europa. eu: Brussels, Belgium, 2016.
[13] European Commission, “Directive 2012/27/eu of the european parliament and of the council of 25 october 2012 on energy efficiency,
amending directives 2009/125/ec and 2010/30/eu and repealing directives 2004/8/ec and 2006/32,” Official Journal, L, vol. 315, pp. 1–56,
2012.
[14] M. Yun and B. Yuxin, “Research on the architecture and key technology
of internet of things (iot) applied on smart grid,” in Advances in Energy
Engineering (ICAEE), 2010 International Conference on. IEEE, 2010,
pp. 69–72.
[15] M. Rana, “Architecture of the internet of energy network: An application
to smart grid communications,” IEEE Access, vol. 5, pp. 4704–4710,
2017.
JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015
[16] D. Niyato, L. Xiao, and P. Wang, “Machine-to-machine communications
for home energy management system in smart grid,” IEEE Communications Magazine, vol. 49, no. 4, 2011.
[17] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, and W. Zhao, “A survey
on internet of things: Architecture, enabling technologies, security and
privacy, and applications,” IEEE Internet of Things Journal, vol. 4, no. 5,
pp. 1125–1142, 2017.
[18] M. Chiang and T. Zhang, “Fog and iot: An overview of research
opportunities,” IEEE Internet of Things Journal, vol. 3, no. 6, pp. 854–
864, 2016.
[19] N. Bui, A. P. Castellani, P. Casari, and M. Zorzi, “The internet of energy:
a web-enabled smart grid system,” IEEE Network, vol. 26, no. 4, 2012.
[20] K. Boroojeni, M. H. Amini, A. Nejadpak, T. Dragičević, S. S. Iyengar,
and F. Blaabjerg, “A novel cloud-based platform for implementation of
oblivious power routing for clusters of microgrids,” IEEE Access, vol. 5,
pp. 607–619, 2017.
[21] H. Tram, “Advanced metering infrastructure as an enabler of demand
response,” in Transmission and Distribution Conference and Exposition,
2010 IEEE PES. IEEE, 2010, pp. 1–1.
[22] Q. Sun, H. Li, Z. Ma, C. Wang, J. Campillo, Q. Zhang, F. Wallin, and
J. Guo, “A comprehensive review of smart energy meters in intelligent
energy networks,” IEEE Internet of Things Journal, vol. 3, no. 4, pp.
464–479, 2016.
[23] E. Patti, A. L. A. Syrri, M. Jahn, P. Mancarella, A. Acquaviva, and
E. Macii, “Distributed software infrastructure for general purpose services in smart grid,” IEEE Transactions on Smart Grid, vol. 7, no. 2,
pp. 1156–1163, 2016.
[24] A. Bahmanyar, S. Jamali, A. Estebsari, E. Pons, E. Bompard, E. Patti,
and A. Acquaviva, “Emerging smart meters in electrical distribution systems: Opportunities and challenges,” in Electrical Engineering (ICEE),
2016 24th Iranian Conference on. IEEE, 2016, pp. 1082–1087.
[25] M. A. Mengistu, A. A. Girmay, C. Camarda, A. Acquaviva, and E. Patti,
“A cloud-based on-line disaggregation algorithm for home appliance
loads,” IEEE Transactions on Smart Grid, 2018.
[26] N. G. Paterakis, O. Erdinc, A. G. Bakirtzis, and J. P. Catalão, “Optimal
household appliances scheduling under day-ahead pricing and loadshaping demand response strategies,” IEEE Transactions on Industrial
Informatics, vol. 11, no. 6, pp. 1509–1519, 2015.
[27] E. Bompard, A. Monti, A. Tenconi, A. Estebsari, T. Huang, E. Pons,
M. Stevic, S. Vaschetto, and S. Vogel, “A multi-site real-time cosimulation platform for the testing of control strategies of distributed
storage and v2g in distribution networks,” in 2016 18th European
Conference on Power Electronics and Applications (EPE’16 ECCE
Europe), Sept 2016, pp. 1–9.
[28] M. Stevic, A. Estebsari, S. Vogel, E. Pons, E. Bompard, M. Masera, and
A. Monti, “A multi-site european framework for real-time co-simulation
of power systems,” IET Generation, Transmission & Distribution, 2017.
[29] M. Stevic, S. Vogel, M. Grigull, A. Monti, A. Estebsari, E. Pons,
T. Huang, and E. Bompard, “Virtual integration of laboratories over long
distance for real-time co-simulation of power systems,” in Industrial
Electronics Society, IECON 2016-42nd Annual Conference of the IEEE.
IEEE, 2016, pp. 6717–6721.
[30] F. Aguirre, A. Silvestris, J. Alemany, and F. Magnago, “A low cost smart
metering infrastructure with centralized demand response management,”
in Innovative Smart Grid Technologies Latin America (ISGT LATAM),
2015 IEEE PES. IEEE, 2015, pp. 520–524.
[31] M. LeMay, R. Nelli, G. Gross, and C. A. Gunter, “An integrated
architecture for demand response communications and control,” in
Hawaii international conference on system sciences, proceedings of the
41st annual. IEEE, 2008, pp. 174–174.
[32] D. Mashima and W.-P. Chen, “Residential demand response system
framework leveraging iot devices,” in Smart Grid Communications
(SmartGridComm), 2016 IEEE International Conference on. IEEE,
2016, pp. 514–520.
[33] H. Kim, Y.-J. Kim, K. Yang, and M. Thottan, “Cloud-based demand
response for smart grid: Architecture and distributed algorithms,” in
Smart Grid Communications (SmartGridComm), 2011 IEEE International Conference on. IEEE, 2011, pp. 398–403.
[34] R. H. Jacobsen, A. G. Azar, and E. S. M. Ebeid, “Design of an eventdriven residential demand response infrastructure,” in Digital System
Design (DSD), 2016 Euromicro Conference on. IEEE, 2016, pp. 38–
45.
[35] B. Bhattarai, M. Levesque, B. Bak-Jensen, J. Pillai, M. Maier, D. Tipper,
and K. Myers, “Design and co-simulation of hierarchical architecture
for demand response control and coordination,” IEEE Transactions on
Industrial Informatics, 2016.
11
[36] S. C. Müller, H. Georg, J. J. Nutaro, E. Widl, Y. Deng, P. Palensky,
M. U. Awais, M. Chenine, M. Kuch, M. Stifter et al., “Interfacing power
system and ict simulators: Challenges, state-of-the-art, and case studies,”
IEEE Transactions on Smart Grid, 2016.
[37] C.-h. Yang, G. Zhabelova, C.-W. Yang, and V. Vyatkin, “Cosimulation
environment for event-driven distributed controls of smart grid,” IEEE
Trans. on Industrial Informatics, vol. 9, no. 3, pp. 1423–1435, 2013.
[38] M. Manbachi, A. Sadu, H. Farhangi, A. Monti, A. Palizban, F. Ponci, and
S. Arzanpour, “Real-time co-simulation platform for smart grid volt-var
optimization using iec 61850,” IEEE Trans. on Industrial Informatics,
vol. 12, no. 4, pp. 1392–1402, Aug 2016.
[39] S. Schütte, S. Scherfke, and M. Tröschel, “Mosaik: A framework for
modular simulation of active components in smart grids,” in Proc. of
SGMS11. IEEE, 2011, pp. 55–60.
[40] S. Rohjans, S. Lehnhoff, S. Schtte, S. Scherfke, and S. Hussain, “mosaik
- a modular platform for the evaluation of agent-based smart grid
control,” in Proc. of PES ISGT, Oct 2013, pp. 1–5.
[41] T. Hess, J. Dickert, and P. Schegner, “Multivariate power flow analyses
for smart grid applications utilizing mosaik,” in Proc. of PES ISGT, Oct
2016, pp. 1–6.
[42] K. Anderson, J. Du, A. Narayan, and A. El Gamal, “Gridspice: A
distributed simulation platform for the smart grid,” IEEE Trans. on
Industrial Informatics, vol. 10, no. 4, pp. 2354–2363, 2014.
[43] M. Brenna, E. De Berardinis, L. D. Carpini, F. Foiadelli, P. Paulon,
P. Petroni, G. Sapienza, G. Scrosati, and D. Zaninelli, “Automatic
distributed voltage control algorithm in smart grids applications,” IEEE
Trans. on Smart Grid, vol. 4, no. 2, pp. 877–885, 2013.
[44] P. Faria, Z. A. Vale, and J. Ferreira, “Demsia demand response simulator
in the context of intensive use of distributed generation,” in Systems Man
and Cybernetics (SMC), 2010 IEEE International Conference on. IEEE,
2010, pp. 2025–2032.
[45] S. Tan, W.-Z. Song, Q. Dong, and L. Tong, “Score: Smart-grid common
open research emulator,” in Smart Grid Communications (SmartGridComm), 2012 IEEE Third International Conference on. IEEE, 2012,
pp. 282–287.
[46] T. K. Wijaya, D. Banerjee, T. Ganu, D. Chakraborty, S. Battacharya,
T. Papaioannou, D. P. Seetharam, and K. Aberer, “Drsim: A cyber
physical simulator for demand response systems,” in Smart Grid Communications (SmartGridComm), 2013 IEEE International Conference
on. IEEE, 2013, pp. 217–222.
[47] M. Pau, E. Patti, L.Barbierato, A. Estebsari, E. Pons, F. Ponci, and
A. Monti, “A cloud-based smart metering infrastructure for distribution
grid services and automation,” Sustainable Energy, Grids and Networks,
2017.
[48] M. Pathirage, S. Perera, I. Kumara, and S. Weerawarana, “A multi-tenant
architecture for business process executions,” in Web services (icws),
2011 ieee international conference on. IEEE, 2011, pp. 121–128.
[49] E. Patti, A. Acquaviva, and E. Macii, “Enable sensor networks interoperability in smart public spaces through a service oriented approach,”
in Advances in Sensors and Interfaces (IWASI), 2013 5th IEEE International Workshop on, June 2013, pp. 2–7.
[50] “Message Queue Telemetry Transport (MQTT),” Available:
http://mqtt.org/.
[51] P. T. Eugster, P. A. Felber, R. Guerraoui, and A.-M. Kermarrec, “The
many faces of publish/subscribe,” ACM CSUR, June 2003.
[52] R. T. Fielding and R. N. Taylor, “Principled design of the modern web
architecture,” ACM Transactions on Internet Technology, vol. 2, no. 2,
pp. 115–150, 2002.
[53] A. Aliberti, C. Camarda, V. Ferro, A. Acquaviva, and E. Patti, “A
participatory design approach for energy-aware mobile app for smart
home monitoring,” pp. 158–165, 2017.
[54] L. Bottaccioli, A. Estebsari, E. Pons, E. Bompard, E. Macii, E. Patti,
and A. Acquaviva, “A flexible distributed infrastructure for real-time cosimulations in smart grids,” IEEE Transactions on Industrial Informatics,
2017.
[55] T. H. Cormen, Introduction to algorithms. MIT press, 2009.
[56] S. Martello and P. Toth, Knapsack Problems: Algorithms and Computer
Implementations. John Wiley & Sons, Inc., 1990.