IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO.
2, FEBRUARY 2015 711
Optimized Multiple Microgrid-Based Clustering
of Active Distribution Systems Considering
Communication and Control Requirements
Seyed Ali Arefifar, Member, IEEE, Yasser Abdel-Rady I. Mohamed, Senior Member, IEEE, and
Tarek El-Fouly
AbstractAn optimized communication and control
infrastructure, based on the microgrid building block
concept, for active distribution systems is essential to
facilitate powerful control framework under the smart grid
paradigm. In this paper, a novel methodology for designing
a communication and control infrastructure is presented.
The new design takes into account both communication
system and distribution system-related aspects. The pro-
posed design facilitates systematic and optimized clus-
tering of the distribution system into a set of virtual
microgrids with optimized communication requirements
while considering the power quality aspects, character-
istics of distributed generation units, distributed energy
storage units, and distributed reactive sources. The new de-
sign facilitates robust infrastructure for smart distribution
systems operation and control, e.g., self-healing control
and optimized system-level operation, by using virtual mi-
crogrids as building blocks in future distribution systems.
The motivations, conceptual design, problem formulation
and solution algorithms are presented in this paper. The
well-known PG&E 69-bus distribution system is selected
as a test case and through several sensitivity studies, the
effect of optimization coefcients on the design, and the ro-
bustness of the algorithm are investigated. Finally, the
algorithm is tested on the IEEE 123-bus distribution system.
Index TermsCommunication system, control center,
graph partitioning, microgrid, power imbalance, Tabu
search (TS).
I. I NTRODUCTION
P OWER engineers nowadays are facing new challenges to
improve the operation of power distribution systems [1].
Introduction of renewable energy resources and energy storage
devices on one hand and the increase in the rate of electricity
consumption, associated costs of large power generating plants
and environmental issues on the other hand, have manifested
the need for new planning, operation and control strategies in
power distribution systems [2][6]. Through the incorporation
Manuscript received October 17, 2013; revised March 15, 2014 and
May 12, 2014; accepted May 24, 2014. Date of publication August 27,
2014; date of current version January 7, 2015.
S. A. Arefifar and Y. A.-R. I. Mohamed are with the Department of
Electrical and Computer Engineering, University of Alberta, Edmonton,
AB T6G 2V4, Canada (e-mail: arefifarr@ieee.org; yasser_rady@ieee.
org).
T. El-Fouly is with CanmetENERGY, Natural Resources Canada
(NRCan), Varennes, QC J3X 1S6, Canada (e-mail: telfouly@nrcan.
gc.ca).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TIE.2014.2352212
0278-0046 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
of new digital and intelligent devices, the power grids are
moving toward a faster-acting, potentially more controllable
and reliable grid, the so-called smart grid [7]. If implemented
and operated properly, the smart distribution system can bring
lots of benefits and advantages to the environment, electric
power suppliers and electricity consumers [8]. For this reason,
they have been under study from several aspects including:
smart grid automation [9], unbalanced optimal power flow [10],
loss minimization [11], distributed-energy-resource manage-
ment [3], control and protection [12], optimal voltage regula-
tion [13], and failure identification [14].
In the context of smart distribution systems, a large number
of emerging devices with wide band of control actions and
options will be adopted. This includes distributed and renew-
able generation units, storage units, custom power devices,
electric vehicle charging stations, demand-side management
controllers, etc. With the increased penetration of these de-
vices along with conventional system devices, the control and
operation of bulk distribution systems can be an issue. A
robust communication and control infrastructure is essential
to facilitate optimized operation of such bulk systems with a
large number of controllable devices and operation scenarios
[15][19]. Two important communication and control strategies
exist for communications and control design in a power system,
namely, they are centralized and decentralized [20], [21].
In the centralized scenario, all the controllable devices receive
the control commands from a central control center. Despite
the advantage of lower costs for installation of communication
and control centers, this option has a higher length of commu-
nication media, relatively larger delays and lower reliability.
The other option is to cluster the controllable devices, such
as distributed generators (DGs), storage devices, controllable
loads, etc., into groups and construct communication zones so
that each zone has its own control center, including voltage and
frequency control units. All the control centers are then super-
vised by a central control center that is at a higher control level
in the designed infrastructure. Through simulations on the IEEE
118-bus test network, it is shown in [21] that having almost the
same total costs for both centralized and decentralized systems,
the latency and reliability in the communication network will
be improved in a decentralized infrastructure.
The question here is what the best scenario is to virtually
split the network into different communication and control
zones and what items should be considered in the design so
that both utility and customers get the maximum benefits.
A good communication infrastructure should be fast, secure,
reliable, and convenient in terms of controlling the devices at
712 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 2, FEBRUARY 2015
different system operating conditions. These parameters and the
installation costs are important and should be considered in an
infrastructure design; however, there are some other issues that
are also worth to be considered in the planning problem. Self-
healing is a relatively new problem in smart distribution sys-
tems, which deals with the algorithms for taking preventative
actions or to handle problems after they occur [22]. The ability
of a distribution network to self-heal in disturbance conditions
is one of the important characteristics of future smart grids
[23]. Under some circumstances, such as disturbances, a self-
healing action is partitioning the system into islands by using
reconfiguration switches so that the healthy part of the sys-
tem remains untouched by the disturbance. Supply-adequacy-
based optimum construction of microgrids in active distribution
systems is addressed in [3], however, communication system
requirements are not considered in the design stage.
Considering the importance of both communication and
self-healing related aspects of system partitioning, this paper
presents a novel methodology to find the optimum communica-
tion and control infrastructure in a smart distribution system.
The proposed design simplifies the control of large systems
with high penetration of DG resources in grid-connected and
islanded modes of operation. It also facilitates systematic and
optimized clustering of the distribution system into a set of
virtual microgrids with optimized communication and control
requirements, while considering the power quality aspects and
characteristics of DG units, distributed energy storage units
(DESR), and distributed reactive sources (DRS). The new
design facilitates robust infrastructure for smart distribution
systems operation and control, e.g., self-healing control and
optimized system-level operation, by using virtual microgrids
as building blocks in future distribution systems. The proposed
design is in line with the recently developed IEEE Std. 1547.6,
which presented a microgrid structure as the building blocks of
active distribution systems [24]. The contributions of this paper
to the research field are the following:
1) development of a systematic strategy for designing a
communication and control infrastructure in smart grids;
2) consideration of communication network costs such as
costs of building a new control center, costs of the com-
munication media and routers in the design;
3) consideration of communication network properties such
as network latency and reliability in the design;
4) taking into account the power-related aspects, such as
power quality issues, characteristics of DG, DESR and
DRS units, in making communication zones as a step
toward an improved self-healing action and coordinated
control actions in a smart grid.
II. M OTIVATIONS AND C ONCEPTS OF D ESIGN
The motivations for having an optimum communication and
control infrastructure for a smart distribution system, as well as
both communications related and power-related aspects of the
design, are explained here.
A. Motivations for the Proposed Design
The problem defined in this paper is a planning problem;
therefore, the audience and user of the proposed approach will
be mostly the utilities. With the high penetration of distributed
resources (DG), the communication infrastructure should cover
key system nodes where DG units, energy storage devices,
major loads, etc., are installed. The utility will use the com-
munication infrastructure for monitoring and control of large
and complicated future distribution systems. In this context,
the ownership of DG does not affect the planning aspects of
the communication system. The communication infrastructure
should cover both customer-owned and utility-owned DG units
to ensure complete access to key generators and loads. When
the infrastructure is designed, both customers and utility will
benefit from it as it facilitates the realization of better control
and operation algorithms for distribution systems. Some of
these benefits are summarized in the following.
1) Distributed Control Strategy: The controllable devices
in a distribution system can be controlled in centralized
or decentralized ways. Other than the improvements in
reliability and latency of the communication network,
in a decentralized approach, the system controllability
becomes easier because of the fewer amounts of control-
lable devices in each control zone.
2) Intelligent Monitoring Options: The number of intelligent
electronic devices for monitoring, protection, and other
purposes such as smart metering, power quality mon-
itoring, and distribution system automation have been
increased in distribution systems, as part of smart grid
projects. Improvements in smart grid communications
system makes the data captured by these new devices
available to multisubscribers and serves multiple intelli-
gent electronic devices infrastructures [25].
3) Advanced Fault Location: In case of faults, the avail-
ability of additional feeder data helps to improve the
accuracy of the fault location methods and subsequently,
improve the self-healing control actions [25], [26]. With
the designed communication and control infrastructure,
more data will be available and as the microgrids are
smaller than the whole system, the fault locating task will
be more accurate and the required self-healing actions
will be done more conveniently.
4) Advanced Self-Healing Control Actions: Self-healing
usually includes system reconfiguration, load shedding
or controlling the dispatchable generators output powers.
Since in the proposed infrastructure, each zone has min-
imum power imbalance and its own control center at the
same time, the self-healing control actions can be done
with minimum required actions and the disconnected
microgrids can operate in autonomous mode with less
amount of load shedding
5) Self-Adequate Microgrids: When the microgrids are con-
structed in a distribution system, the less the generation-
load imbalance within them will result in the more
self-sufficient microgrids. Thus, more loads can be sup-
plied in case of autonomous-mode operation of micro-
grids in the distribution system, and the energy losses on
power lines connecting the microgrids will be minimized.
6) More Reliable Distribution Systems: Reliability is an
important issue in designing current distribution systems.
With the proposed design, the system will be split into
AREFIFAR et al.: CLUSTERING OF SYSTEMS CONSIDERING COMMUNICATION AND CONTROL REQUIREMENTS
several self-adequate microgrids. Other than easing the
self-healing control actions, which increase the reliability
of the system, the self-sufficiency of microgrids will
also improve the reliability indexes such as the System
Average Interruption Frequency Index (SAIFI), System
Average Interruption Duration Index (SAIDI), etc.
The aforementioned items present some justifications and
motivations behind the proposed communication and control
infrastructure design strategy. Some specific aspects of the
design such as communications- and power-related aspects are
explained with more details in the following sections.
B. Communication and Control-Related Aspects
The communication network is an essential part of the power
system as a failure in communication may affect the operation
of the electrical system itself. Furthermore, an optimized com-
munication infrastructure is essential to facilitate smart grid
control and operation functions, such as self-healing control
and distributed energy management. Thus, the communication
infrastructure will be as important as the electrical infras-
tructure [27]. The communication network performs the data
exchange between data resources, control centers, and control-
lable devices; therefore, it will be a vital part of wide-area mea-
surement systems, wide-area control systems, and wide-area
protection systems. The base-line communication and control
infrastructure is usually designed for long-term purposes, and
it may evolve depending on new system-level needs. At the
base-line planning stage, however, there are some issues that
should be considered. In case of future needs for upgrading
the communication system, the optimum design should be
updated accordingly. Key long-term objectives are considered
in this paper as a typical base-line communication and control
infrastructure design for smart and active distribution systems.
Three important factors are cost-related issues, latency, and
reliability of the network. The cost-related issues include the
costs of communication media, communicating routers and
communication and voltage and frequency control centers and
the latency/reliability of the communication system is mainly
based on the latency/reliability of communication media and
the communicating routers. Therefore, the optimum infras-
tructure is the one with optimum number of communication
zones and minimal total communication costs. In spite of the
minimized costs, the system latency should be kept less than a
threshold, and the minimum required system reliability should
be maintained.
C. Power-Related Aspects
In response to disturbances, a self-healing system reconfig-
uration that splits a power network into self-sufficient islands
can stop the propagation of disturbances and avoid cascading
events. Traditionally, only the active power has been considered
in system partitioning [28], [29]. However, by considering
the reactive power as an important role player in supporting
the voltage profile and knowing that a significant mismatch
of reactive power supply and demand causes high- or low-
voltage conditions within islands, both real and reactive power
713
balance can be considered in system partitioning [30]. Although
the controllable generation and load must be balanced once
a system partitions, if we consider minimizing the power
imbalance in the zones and in steady-state conditions at the
planning stage, balancing generation and load considering the
system dynamics will be much more convenient and effective in
the operating stage. Therefore, an optimum partitioned system
is the one with minimum generation-load imbalance in each
zone. With the proposed design in this paper, each island can
behave as a microgrid and may operate in grid connected or
islanded mode, with minimum interactions from other parts of
the system. Moreover, the power balancing in each microgrid
will be done easily, as each microgrid has its own control
center. It should be noted that since the research in this paper
is a long-term planning problem, the system transients and
dynamics, which may occur during islanding process and do
not last for a long period, are not considered here. To account
for special distribution system characteristics associated with
power balancing, such as the intermittent nature of DG units
and loads, a probabilistic power balancing approach is proposed
in this paper.
D. Combination of the Two Aspects
Both communication and control and power-related aspects
are taken into account for finding the optimum communication
and control infrastructure. With this design, from the commu-
nication and control point of view, the whole system can be
controlled more conveniently in grid-connected and islanded
mode of operation by using decentralized control centers to
optimize the overall operational aspects of the system. From
the power point of view, clustering the system into self-
sufficient microgrids, regardless of grid-connected or islanded
mode operation, will ease the self-healing control actions such
as system reconfigurations, load shedding or controlling the
DGs outputs in case of disturbances. One obvious advantage
of this formulation is that if for any reason, the system is
split into zones, each zone has the minimum generation-load
imbalance and, at the same time, has its own control center.
In this scenario, the preventive actions in the disturbed zones
and healthy zones can be performed more conveniently and
more intelligently. Although in some zones the generation level
may not meet the required consumption at the peak load level,
the islanding may be possible at light-load periods or through
minimum load shedding actions, and this will be beneficial for
both utility and customers. Furthermore, the created virtual mi-
crogrids can be controlled in a coordinated manner to optimize
the overall system-level operational aspects. In other words,
the extended formulation proposed in this paper provides a
flexible infrastructure for the distribution system operation by
applying the concept of virtual microgrids. The main design
concepts proposed in this paper and how the virtual microgrids
are clustered are shown in Fig. 1.
III. M ODELING OF L OADS AND DGs
To obtain an optimum communication and control infras-
tructure in the presence of the uncertain nature of DG units
and loads, the characteristics of loads and DG units should be
714 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 2, FEBRUARY 2015

power of PV modules can be easily generated. The Beta PDF

used to describe the probabilistic nature of solar irradiance, is
as shown in
( + )
fb (s) = s(1)
()()
(1 s)(1) 0 s 1, 0, 0 (1)

where s is solar irradiance (kW/m2 ), fb (s) is Beta distribution

function of s, and and are parameters of the Beta distribu-
tion function. The parameters of Beta distribution function are
calculated by using the mean () and standard deviation () of
random variable s, as shown in
 
(1 + )
= (1 ) 1 & = . (2)
2 1
The Beta PDF is generated for each time segment (one hour
for this research) and the probabilistic output power of the PV
modules is then calculated using the following equations [31]:
 
Fig. 1. Main design concepts and the virtual microgrids. NOT 20
Tcy = TA + say (3)
0.8
considered in the design. For this purpose, DG units are mod- Iy = say [Isc + Ki (Tc 25)] (4)
eled as typical combination of three types of DGs, including
Vy = Voc Kv Tcy (5)
photovoltaic (PV) modules, wind turbines, and biomass gen-
P sy (say ) = N F F Vy Iy (6)
erators. The nature of renewable resources are probabilistic;
VMPP IMPP
therefore, the solar irradiance and wind speed for each hour of FF = (7)
the day are modeled by Beta and Weibull probability density Voc Isc
functions (PDFs) by using historical data [31]. To get the PDFs, where Tcy and TA are cell and ambient temperature during
the study can be done over a selected period of one year, and state y ( C); Kv and Ki are voltage and current temperature
four days can be selected as representatives of four seasons. coefficients (V / C and A/ C); NOT is nominal operating
The day representing each season is further divided into 24-h temperature of cell per C; F F is fill factor; Isc and Voc are
time segments, each having a PDF for solar irradiance and short circuit current (A) and open circuit voltage (V ); N is the
wind speed. The load is also modeled as a firm generation number of cells; IMPP and VMPP are current (A) and voltage
using IEEE-RTS [32]. In order to integrate the output power (V ) at maximum power point; P sy is the output power of
of PV modules and wind turbines as multistate variables in the PV module during state y, and say is the average solar
the formulations, the continuous PDF of each is divided into irradiance of state y.
different states. The selected number of states affects the accu-
racy and complexity of the formulation. In this research, the
B. Wind Turbine Generators
output power of the wind turbine and PV modules for each
hour of the day is divided into 12 segments. This means for The output power of wind turbines depends on the wind
each hour of the day, we have 12 states of solar DG power and speed and parameters of the wind-power conversion curve.
12 states of wind DG power with different probabilities. Similar to the PV module generators, once the wind speed is
Assuming that solar irradiance and wind speed states are in- modeled for each hour by using a PDF, the probabilistic func-
dependent, the probability of any combination of the load and tion of the output power of wind turbines can be generated eas-
generation is obtained by convolving the two probabilities. ily. The wind speed is usually modeled using the Rayleigh PDF,
Therefore, for each hour, we have 144 (1212) number of which is a special case of the Weibull PDF, as shown in (8) [31]
states with different probabilities and for each day there are      
2v v 2
3456 (24144) states. To obtain the power and energy losses f (v) = 2
exp (8)
c c
for the system, the deterministic power flow is run for each
state, and the results are accumulated considering the proba- where c is called the scale index. The mean value of wind speed
bility of each state. (v) is calculated using the historical data for each time segment
and then the scale index can be calculated, as shown in
A. PV Module Generators        
2v v 2
vm = vf (v)dv = v 2
exp dv
The output of each PV module depends on the amount of c c
0 0
solar irradiance, ambient temperature and characteristics of the
module itself. When the solar irradiance is modeled for each
= c, c 1.128vm . (9)
hour by using a PDF, the probabilistic function of the output 2
AREFIFAR et al.: CLUSTERING OF SYSTEMS CONSIDERING COMMUNICATION AND CONTROL REQUIREMENTS
The output power of a wind turbine is calculated using
0, 0 vaw vci , vco vaw
vci )
P vw (vaw ) = Prated (vvaw , vci vaw vr
r vci
Prated , vr vaw vco
(10)
where vci , vr , and vco are the cut-in, rated, and cut-out speed
of wind turbine, respectively; P vw is the output power of wind
turbine during state w; and vaw is the average wind speed of
state w.
C. Biomass Generators
The biomass generators are assumed to be firm generators
with constant output powers without uncertainty. In case of
uncertainty in the factors that affect the availability of biomass,
such as rain, the biomass DGs output powers can be modeled
in a similar way that has been presented for PV module or
wind turbine generators by using the probability function of
availability of the biomass for the planning period.
D. DESRs and DRSs
The DESRs are modeled as loads during the charging period
and as generators during discharging period. It is also assumed
that the DESRs charge only during off-peak hours of the day
715
where LoMC is the total length of communication media,
N oRC is the total number of routers, and N oZ is the number
of communication or control zones. The coefficients , , and
are the cost of communication media, cost of each router, and
cost of installing a communication and control center (including
voltage and frequency controller units). Their dimensions are
$/km, $/router and $/(control center), respectively. Minimizing
F1 will minimize the total cost of communication infrastructure
by finding the communication-based virtual microgrids.
The other objective is to minimize the real and reactive power
imbalance between generation and loads within the zones. This
is a step toward 1) having a self-healing strategy for system
reconfiguration that splits the power system into self-sufficient
islands in response to disturbances; and 2) creating virtual
microgrids where controllable devices that affect the power
flow (e.g., DG units, demand-side management, etc.) can be
clustered. This objective function can be formulated as
a Pindex + b Qindex
F2 = (13)
(N oZ 1)
where Pindex and Qindex are probabilistic indexes to represent
the probabilistic real and reactive power of the virtual cut set
lines and are calculated, as shown in (14). The denominator is
set to normalize the cost function

N 
N
and discharge only during the on-peak hours. Therefore, for Pindex = |Pijk |k & Qindex = |Qijk |k (14)
each period of 24 h, there is one charge-discharge cycle. A con- k=1 k=1
straint has been introduced on the amount of power generated
by the DESRs, as shown in where i and j are the nodes, which connect two zones to-
 
PDESR_i,t PLoad_j,t t (11)
i j
where Pload_j,t is the real power consumed at bus j, time t and
PDESR_i,t is the real power generated by DESRs at bus i and
time t. This constraint keeps the output power of DESRs always
less than the total demand during each hour and guarantees that
the reverse power flow is coming from the DG units during any
hour. The DRSs are also modeled as simple reactive sources
with specific rated capacity.
IV. P ROBLEM F ORMULATION
Optimization problems are constructed from an objective
function and some constraints. This section explains the pro-
posed problem formulation and associated constraints.
A. Optimization Objective Function
Two different objective functions are considered in this re-
search, namely, they are the communication and control re-
lated and the power balance related. The communication and
control cost function to be minimized includes the cost of
installation of new control centers, which includes voltage and
frequency controller units; cost of network switches and routers
and the cost for length of communication media that can be
represented as
F1 = LoMC + N oRC + N oZ (12)
gether; Pij _k and Qij _k are the real and reactive power flowing
between buses i and j at the generation-load state k; N is
the number of states for each year (4(seasons) 90(days)
3456(states/day)); and k is the probability of the state, in
which the real and reactive powers are calculated. The power
balance objective function can be calculated by performing a
probabilistic power flow in the system. Since the two aspects
are to be considered simultaneously, the final objective function
(F ) can be represented as the weighted summation of F1 and
F2 , as shown in
M in(F ), F = K1 F1 + K2 F2 . (15)
Based on the definition of , , and , the dimension of F1
is costs in dollars, if we determine the cost of interruption due
to the supply inadequacy or power quality issues of microgrids
(which is out of the scope of this paper), then we can define
K2 as dollar per imbalance ratio. With such definitions the
objective function of the problem will be in dollars.
B. Optimization Constraints
There are two types of constraints for the optimization prob-
lem, communication related and power related. From a commu-
nication point of view, two constraints, i.e., network latency and
reliability, should be satisfied for the optimum infrastructure.
Although the network latency is not only a function of the
length of media and the number of routers in the system;
however, these two indexes provide sufficient information for
716 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 2, FEBRUARY 2015
the design regarding the systems latency. On the other hand,
since the power systems typically enjoy relatively low data rate
communications, other factors (e.g., performance loss, band-
width cost) can be safely ignored. It is noted that by assuming
a proper medium access control for low-rate communications,
the operational delay can have a negligible effect on the latency.
Therefore, the total systems latency is calculated, as shown in
(16) [21]
T = TSerial + Tbetween.packet + Tpropagation + Trouting . (16)
Tserial and Tbetween.packet are serial delay and between packet
delay, which are constant and Tpropagation and Trouting are
propagation delay and routing delay and are calculated, as
shown in (17); therefore, the total network latency will be
T = Constant + Tpropagation + Trouting
= Constant + LoMT /v + N oRT Trouter (17)
where v is the velocity at which the media is sent through
(close to speed of light), LoMT is the maximum length of
communication media from control center, N oRT is number
of routers seen from control center to the farthest node, and
Trouter is the router delay. For the optimum infrastructure, we
should have
T Tmax . (18)
The second constraint is the threshold for the network relia-
bility. The communication reliability is function of number of
routers and length of media. Each node is in service only if all
the links and paths related to this node work properly; therefore,
the communication networks reliability in each zone can be
formulated as
Rz = (Rrouter )N oRR RLoMR (19)
where Rrouter is the reliability of each router, RLoMR is the
reliability of total length of communication media from control
center to each node, and N oRR is the total number of routers in
the zone. Since the network reliability in each zone should be
higher than a threshold, the minimum calculated reliability for
each zone is considered as the reliability index (R). Here, also
for the optimum infrastructure, we should have
R Rmin . (20)
From power distribution system point of view, one necessary
condition for an island to be successful is
PDG PLoad + PLoss (21)
where PDG is the generated power of the DGs within the
microgrid, PLoad is the load power of the island, and PLoss
is the power losses of the island and assumed to be 5% of
the current load [33]. On the other hand, according to [34]
for an islanding operation of microgrid, the penetration level
of dispatchable DG units should be at least 60% of the total
peak load demand at the time of islanding. For this research,
it assumed that biomass DG units are dispatchable; therefore,
the second necessary condition for a successful operation of the
microgrid is
PBM 0.6 PDG (22)
where PBM is the generated power of biomass DGs in the mi-
crogrid. Equations (21) and (22) are two necessary conditions
for the successful operation of microgrids and can be satisfied
by shedding the loads according to their requested reliability.
In this research, the load shedding starts from smaller loads to
larger ones until the conditions are satisfied. Moreover, some
other constraints should be satisfied to guarantee the security
of supply for the constructed microgrids. They are the voltage
magnitude (V ) and frequency (f ) limits on all system buses
in the microgrids, and maximum lines currents (I), which are
formulated as follows:
Vmin Vi Vmax i
fmin f fmax (23)
and for power line current limits
Ii Imax,i i. (24)
The optimum microgrid design should satisfy these conditions
in an islanded mode of operation. In Section VI, sensitivity
studies are conducted to assess the effects of changing the
coefficients on the optimum design.
V. S OLUTION A LGORITHMS
The formulated problem in Section IV is a comprehensive
problem with combinatorial nature and demands efficient solu-
tion algorithms. This problem has two different objective func-
tions, which are aggregated into a single-objective function, by
using weighting coefficients. It is assumed that the weighting
coefficients such as K1 and K2 are predetermined based on
economic and system-related studies; therefore, the problem
can be seen as a single-objective optimization problem. Heuris-
tic optimization techniques are well-suited for such optimiza-
tion problems. In this paper, we have used three different
types of algorithms at different stages, including Tabu search
(TS) optimization algorithm, as the main optimization method,
graph theory related techniques for suboptimization problems
and NewtonRaphson-based probabilistic power flow methods.
Several heuristic optimization techniques other than TS have
been proposed in literature for graph partitioning, which cope
with the combinatorial nature of the present problem, such as
those mentioned in [27], [28]. Since the algorithm proposed by
the authors in this paper is effective and sufficient for solving
the defined problem, and the main contribution of this paper
is the new problem formulation that combines communication
and power balance aspects, other methods are not considered
in this research for splitting the network. The flowchart of
the algorithms is shown in Fig. 2. The optimization process
for selecting the optimum cut set lines are explained in the
following.
AREFIFAR et al.: CLUSTERING OF SYSTEMS CONSIDERING COMMUNICATION AND CONTROL REQUIREMENTS
Fig. 2. Flowchart of the algorithms used for the design.
A. Starting the Optimization Process
The optimization process starts with finding a feasible so-
lution as a starting point, searching within the neighbors and
continues until a certain criterion, which is usually the maxi-
mum iteration numbers, is reached. A feasible solution is an
infrastructure constructed through virtually cutting some power
lines in the network. Each feasible solution can be represented
by a vector (G), as shown in
G = [1 0 1 1 0 . . . . . . 1 0 1 1] (25)
where the length of the vector is equal to the number of power
lines, which can be considered as cutting edges to construct the
zones. Each component of the G vector represents one physical
line. The number 1 and 0 for each component of G vector
represents an inside or between microgrids line, respectively.
The next step is to make a set of neighbors for the starting
point. A neighbor can be defined in many ways; here, we have
selected each neighbor by changing some components (e.g.,
five components) of the G from 0 to 1 or vice versa and previous moves or the moves that resulted in the optimum point
checking the feasibility of the resulted infrastructure in terms
of the constraints. The constraints should be satisfied for all
neighbors as explained in the next part.
717
i.e., latency and reliability. The latency-related constraints are
checked by calculating LoMT and N oRT . These values are
calculated by solving the minimum spanning tree (MST) [35]
and the shortest path [36] algorithms. By finding the MST
in each zone, the structure of the communication network is
determined. Afterward, the shortest path from each bus to the
other buses in the same microgrid is calculated. The bus that has
the shortest length of communication media to all other buses
is selected as the control center. After determining the control
center, the LoMT and N oRT can be calculated in each zone,
and the maximum values calculated for the zones will be used
to check the feasibility of infrastructure in terms of network
latency (less than 100 ms).
The reliability constraint is checked by calculating LoMR
and N oRR . Through the MST problem solved for the latency
constraint, these two parameters can be calculated for each
zone, and the minimum value for the zones will be used to
check the feasibility of the infrastructure in terms of network
reliability (more than 0.99). The power-related constraints can
be checked by performing power flow in the constructed micro-
grids. Each G vector represents a set of virtual cut set lines that
construct a microgrid structure. For a structure to be acceptable,
all the three constraints, voltage limits on all system buses in
the microgrids, system frequency, and power line current limits,
should be satisfied. The next step is to calculate the objective
function and select the best point among the neighbors.
C. Selection of the Best Point
The objective function includes two parts, i.e., F1 and F2 . F1
is determined as a weighted summation of LoMC , N oRC , and
N oZ, which are determined from solving the suboptimization
problem. F2 is related to the real and reactive power flowing
in each cut set line of the system and can be determined once
at the beginning through probabilistic power flow calculations
considering typical characteristics of DG units and loads in
active distribution systems. After calculating the objective func-
tion for all the points (G vectors) in the neighborhood, the
best neighbor (vector) will be set as the new starting point and
the algorithm continues until reaching the maximum iteration
number. Finally, the optimum result is selected as the best point
(vector) of each set of neighbors.
D. Implementing Memory Structures
To avoid stopping over a local optimum, and to prevent
cycling around it, the Tabu List (TL) is introduced, which
prevents to visit the best solutions that have been visited in
in previous regions. To construct the TL from the best recently
visited solutions, a unique quantity for each vector G is used,
which is calculated in

n
B. Checking the Optimization Constraints G(i) 2i . (26)
i=1
As explained in Section IV-B, two types of constraints
should be satisfied, the communication related and power re- Furthermore, the TS algorithm uses different memory struc-
lated. The communication-related constraints include two parts, tures to avoid random search. In this paper, the short-term
718 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 2, FEBRUARY 2015
Fig. 3. The 69-bus distribution system with DGs locations.
memory (STM) and the long-term memory (LTM), are imple-
mented. The STM memorizes the common features of subopti-
mal solutions for a number of iterations and tries to search for
the solutions with similar features in that region. The LTM is
used to diversify the search by jumping to new regions to find
the global optimum. This long-term memory will keep track of
the common features of all initial starting points in different
regions to avoid restarting from similar previously used starting
points. The two types of memories have been implemented by
using two different vectors with the same length as G.
In the next sections, the proposed algorithms have been
applied to a 69-bus distribution system. After several sensitivity
studies for the formulation and robustness of the algorithm, the
algorithm has been applied to a larger system, the IEEE 123-bus
distribution system, and the results are presented.
VI. I MPLEMENTATION AND S ENSITIVITY S TUDIES
The well-known PG&E 69-bus distribution system [37] is
selected as the test system for implementation of the algorithm
and sensitivity studies. The locations of DGs, DESRs, and
DRSs are arbitrary selected to be as shown in Fig. 3. The total
rated capacities of DG units are 250, 150, and 600 kW, for
wind turbines, PV modules and biomass DGs, respectively. The
total rated power of DESRs and DRSs are also 450 kW and
400 kVAr, respectively, in 25 kW/kVAr and 50 kW/kVAr units.
The location of reconfiguration switches are also shown in
Fig. 3 [38]. In the next sections, the optimum communication
and control infrastructure are designed for this system consider-
ing different objective functions. The goal or objective function
defined in this paper can represent real costs. The coefficients a,
b, , , , K1 , and K2 will be different for different distribution
systems depending on the economic conditions. How variation
of these coefficients affects the optimum design for communi-
cation and control infrastructure is investigated here.
For the sensitivity studies, the cost function for all cases is
normalized by using one of them as a reference (arbitrary) and
the rest is shown as percentages of this value. This will help
to have a more clear evaluation of the effect of varying the
coefficients on total costs.
TABLE I
V IRTUAL C UT S ETS C ONSIDERING P OWER R ELATED A SPECTS
A. Power Imbalance Related Objective Function Only
In this part, we only consider the power imbalance related
objective function. In other words, K1 = 0 & K2 = 1. Math-
ematically, the optimum result will occur when the system is
split into two virtual microgrids and the connecting power line
carries the minimum power. In order to have more feasible
values (more than two microgrids), another constraint is added,
at this stage only, which is on the number of zones to be more
than 5. For the rest of simulations in this paper, there is no limit
on the number of microgrids, and the optimum number will
be determined as a result of solving the optimization problem.
Table I shows the optimum virtual cut set lines to construct the
microgrids. In Table I, the cut set line j is the line between
bus j and j + 1, line jei is the ith branch starting from bus
j. The sensitivity studies at this stage, reveals that the system
partitioning results will be different based on selection of real
and reactive power indexes, or both, for minimizing the power
imbalance in the zones. For this table, the goal is normalized
using the case that has a = b = 0.5, and the rest is shown as
percentage values of this case.
B. Communication-Related Objective Function Only
In this part, we only consider the communication-related
objective function and its constraints to investigate the impact
of communication-related coefficients on the optimum design.
In other words, K1 = 1 & K2 = 0. The results of sensitivity
studies are shown in Table II, where , and represent the
cost of length of media, cost of a router, and cost of building
a control center, respectively. Since the cost of length of media
per kilometer and the cost of a communication router are in the
same range and very different from the cost of installation of
AREFIFAR et al.: CLUSTERING OF SYSTEMS CONSIDERING COMMUNICATION AND CONTROL REQUIREMENTS
TABLE II
V IRTUAL C UT S ETS C ONSIDERING C OMMUNICATION A SPECTS
TABLE III
O PTIMUM R ESULTS C ONSIDERING B OTH C OMMUNICATION AND
P OWER R ELATED A SPECTS
a new control center, and are assumed to be constant (0.5
each) and has been changed to see its effect on the optimum
communication and control infrastructure. It is seen that the
results are very dependent on . By changing it from 50 to 500,
the total infrastructure changes from 9 control zones to 3 control
zones. The value of varies for different systems and should
be set based on the cost of building the communication center
in each area, its maintenance, etc. For this table, the objective
function is normalized using the case that has = = 0.5 and
= 200.
C. Considering Both Objective Functions
Both power imbalance and communication cost-related ob-
jective functions are considered here. For this purpose, the al-
gorithm is applied under different circumstances with different
optimization coefficients, and the optimum communication and
control infrastructure are developed as presented in Table III.
For the sensitivity studies here, it is assumed that a and b
are equal to 0.5, and , , and are 0.5, 0.5, and 200,
respectively. Table III shows that depending on the importance
of communication related or power imbalance related aspects,
the optimum communication and control infrastructure will be
different. For Table III and the rest of tables in this paper,
the objective function is normalized using the case that has
a = b = = = 0.5, = 200, K1 = 0.7, and K2 = 0.3. As
explained before, the selection of these parameters for a refer-
ence case is arbitrary, and normalizing is done for having a more
clear comparison between the objective functions of different
cases. The goal in Table III has large variations for different
cases. The reason is that K1 and K2 , which are determined
based on circumstances of communication- and power-related
aspects of the problem, have significant impacts on the total
objective function. The optimum infrastructure for the case that
K1 and K2 are 0.7 and 0.3, respectively, is shown in Fig. 4.
The optimum infrastructure in this case, divides the system
into 5 zones with each having different controlled buses. As
719
mentioned before, it is assumed that each bus in the system is
equipped with a controllable device such as distributed and re-
newable generation units, storage units, custom power devices,
electric vehicle charging stations, demand-side management
controllers, etc.
The sensitivity studies presented here reveals that the de-
signed communication and control infrastructure is dependent
on the optimization coefficients and for having an optimum
communication and control infrastructure, all the optimization
coefficients should be selected properly.
VII. R OBUSTNESS OF THE A LGORITHM
The research presented in this paper is related to the plan-
ning stage and the audiences of this paper are mostly util-
ities. From the utility point of view, a communication and
control infrastructure or any other long-term plan will be
acceptable if it is optimal or near optimal, in case of vari-
ation of system conditions, for the designated long-term pe-
riod. In this research, the sensitivity of the proposed design
to variation of some important characteristics of a distribu-
tion system that may change over time is investigated. These
characteristics include variation of loads, penetration level
of DGs on some or all system buses, adding new DGs to
random buses, and variation of system topology. How the
variation of each parameter will affect the optimum infras-
tructure and whether the planned infrastructure is still op-
timal or near optimal design are investigated here. For this
purpose, several case studies are presented; first, the effect of
changing the loads and generation levels, then the effect of ad-
ding DGs to some random buses, and finally the effect of system
reconfiguration on the optimum design is observed. For the
sensitivity studies here, any of the cases mentioned in Table III
can be used. As a sample case, here, it is assumed that a
and b are equal to 0.5, and , and are 0.5, 0.5, and 200
respectively, also K1 and K2 are 0.7 and 0.3, respectively.
A. Load and Generation Levels
The load level or consumers of electric power in the system
usually increases during a long period of time. The penetration
level of DG units in the system also varies during a long-
term operating period depending on several factors such as
economic, environmental, and weather conditions, based on
distribution system companies decision. This part examines
the effect of long-term variations in the load and generation on
the optimum design. For this purpose, three different scenarios
are investigated. First, it is assumed that the load and generation
rated powers are changed from 90% to 150% for all system
buses, and the results are shown in Table IV. In the second
scenario, only the loads on some buses of the system are
changed from 90% to 150% (even buses) and the rest of loads
and DGs penetration are kept constant. The results related to
this part are shown in Table V. Finally, the penetration levels
of DGs on some system buses (even buses) are changed from
90% to 150% and the rest of DGs generations and load levels
are kept constant, as shown in Table VI. The sensitivity studies
here show a little difference in the optimum infrastructure in
each case; however, from a communication and control point
720
Fig. 4. Optimum communication and control infrastructure for 69-bus system.
TABLE IV
O PTIMUM V IRTUAL C UT S ETS IN C ASE OF VARIATION OF
L OAD AND G ENERATION L EVELS
TABLE V
O PTIMUM V IRTUAL C UT S ETS IN C ASE OF VARIATION OF
L OAD L EVELS ON E VEN B USES
of view, the optimum infrastructure is very similar. This means
that if for any reason the load or generation changed over a
long period of time for all or some of the system buses, the
designed infrastructure is still valid and close to optima in
terms of communications and power-related aspects. It is clear
that further increase in the load or penetration level of DGs
will affect the optimum design. In such cases, the optimum
communication and control infrastructure should be updated
accordingly.
B. Adding DGs to Random Buses of the System
During a long-term period, the total number of DGs may in-
crease in the system based on utilities or DG owners decision.
The impact of adding DGs to the system buses on the optimum
communication and control design is investigated here. For
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 2, FEBRUARY 2015
TABLE VI
O PTIMUM V IRTUAL C UT S ETS IN C ASE OF VARIATION OF
G ENERATION L EVELS ON E VEN B USES
TABLE VII
O PTIMUM V IRTUAL C UT S ETS W HEN A DDING DG S TO R ANDOM B USES
this purpose, it is assumed that different types of DG units,
including wind turbines, photovoltaic modules and biomass
generators are randomly added to some of the system buses,
and the optimum communication and control infrastructure is
designed and compared for the updated systems. The results
related to this part are shown in Table VII. In this table, the
total penetration level of DGs is increased from 0 to 350 kW in
25 kW steps. The second column, WSB, represents the number
of wind, solar and biomass DG units added to some random
buses of the system. For instance, WSB = 325 means three
units of 25 kW wind turbines, two units of 25 kW photovoltaic
module and five units of 25 kW biomass generators are added
to ten (3 + 2 + 5) random buses of the system. The sensitivity
studies here reveal that the optimum designed infrastructure
is similar in all cases. This means that adding up to 350 kW
new DGs to the system (%35 of existing DGs capacities) does
not have significant impact on the location of sectionalizing
AREFIFAR et al.: CLUSTERING OF SYSTEMS CONSIDERING COMMUNICATION AND CONTROL REQUIREMENTS
TABLE VIII
O PTIMUM V IRTUAL C UT S ETS IN C ASE OF VARIATION
OF S YSTEM C ONFIGURATION
switches or optimum communication and control infrastruc-
ture. It should be noted that adding more DGs to the system
may affect the optimum infrastructure, and in such cases, the
communication and control infrastructure should be modified
accordingly.
C. Distribution System Reconguration
The system configuration might change through operation of
sectionalizing switches. In such cases, whether the designed
communication-control infrastructure is still valid and close
to optima should be investigated. In Table VIII, the system
configuration is changed and the optimum infrastructure has
been designed for the system. L-in and L-out in the table be able to feed the loads in the zone during light-load periods.
represent the connected and disconnected line, respectively.
Table VIII reveals that the optimum designed communication
and control infrastructures under different system topologies
are similar. This means that if for any reason the system
configuration changed over a period of time, the designed
infrastructure is still acceptable. Similarly, in this case, it is
clear that major topological changes in the system configuration
may affect the optimum design, and in such cases, the optimum
communication and control infrastructure should be updated
accordingly.
VIII. L ARGE S YSTEM A PPLICATIONS
The proposed algorithm was tested on the adopted IEEE
123-bus distribution system and the results are presented here.
For this system also, the locations of DGs, DESRs, and DRSs
are arbitrary selected to be as shown in Fig. 5. The total
rated capacities of DG units are 350, 250, and 800 kW, for
wind turbines, PV modules and biomass DGs, respectively. The
total rated power of DESRs and DRSs are also 800 kW and
700 kVAr, respectively, 50 kW/kVAr units. The locations of
DGs in the system are shown in Fig. 5. In order to design
the optimum communication and control infrastructure for the
123-bus distribution system, the optimization parameters are
selected the same as in Section VII. The processing time
for 1000 iterations on an Intel Core2 Duo with CPU
E7500@2.93GHz for 69-bus system was 24 min and 46 s and
for the 123-bus was 38 min and 53 s. Since this is a planning
not operation problem, there is no limitation for the processing
time; therefore, to get higher quality results, we can increase the
number of neighborhood and iterations in the TS algorithm. The
optimum communication and control infrastructure designed
for this system is shown in Fig. 5. The 123-bus distribution
721
Fig. 5. Optimum design for IEEE 123-bus distribution system.
system is split into 7 virtual zones. In each zone, there are some
DGs, DESRs and DRSs. Through the optimum design, every
zone has minimum real and reactive power imbalance between
generation and load; therefore, the DGs will be able to feed
the zones with minimum load shedding actions or even may
The total cost of communication and control infrastructure is
also minimized while maintaining the acceptable reliability and
latency of the communication network. With such design for
the communication and control infrastructure, each zone has its
own control center. Having a control center and communication
network, with acceptable reliability and latency in each zone,
will ease the operation and control of generation and loads
such as controlling the DGs output powers, load shedding and
performing demand-side energy management. Furthermore, in
case of faults, with the minimal self-healing actions, such as
load shedding in the zones, the maximum number of loads can
still get their service from utility.
IX. C ONCLUSION
This paper has presented a strategy for designing a com-
munication and control infrastructure in a distribution system
based on the virtual microgrid concept. The new design takes
into account both communication and power-related objective
functions. In the optimum infrastructure, the total costs related
to building communication and control infrastructure should be
minimized while maintaining the acceptable network latency
and reliability. On the other hand, the optimum partitioning
scenario in case of emergency is the one with minimum power
imbalance in each zone. This will ease the control strategy in
each island and protects other healthy areas of the network
from the disturbances as a self-healing action in a distribution
system. The sensitivity studies show that the optimum infras-
tructure depends on these coefficients, which differ for different
systems. The robustness of the infrastructure to the variation
of loads and DG penetration levels, adding new DGs to the
system, and system reconfiguration is investigated. The results
show that the final design is not very sensitive to the DG and
722 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 2, FEBRUARY 2015
load level, adding new DGs and to the system topology. The
algorithms are then applied for the IEEE 123-bus distribution
system and the results are presented. By considering two dif-
ferent concerns of distribution system engineers, the proposed
novel strategy for designing the optimum communication and
control infrastructure is a major step toward having a more
intelligent and cost efficient smart grid.
R EFERENCES
[1] G. T. Heydt, The next generation of power distribution systems, IEEE
Trans. Smart Grid, vol. 1, no. 3, pp. 225235, Dec. 2010.
[2] S. A. Arefifar and Y. A.-R. I. Mohamed, Probabilistic optimal reactive
power planning in distribution systems with renewable resources in grid-
connected and islanded modes, IEEE Trans. Ind. Electron., vol. 61,
no. 11, pp. 58305839, Nov. 2014.
[3] S. A. Arefifar, Y. A.-R. I. Mohamed, and T. El-Fouly, Supply-adequacy-
based optimum construction of microgrids in smart distribution systems,
IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 14911502, Sep. 2012.
[4] N. Jain, S. Singh, and S. Srivastava, A generalized approach for DG
planning and viability analysis under market scenario, IEEE Trans. Ind.
Electron., vol. 60, no. 11, pp. 50755085, Nov. 2013.
[5] F. Pilo, G. Pisano, and G. G. Soma, Optimal coordination of energy
resources with a two-stage online active management, IEEE Trans. Ind.
Electron., vol. 58, no. 10, pp. 45264537, Oct. 2011.
[6] S. Paudyal, C. A. Canizares, and K. Bhattacharya, Optimal operation of
distribution feeders in smart grids, IEEE Trans. Ind. Electron., vol. 58,
no. 10, pp. 44954503, Oct. 2011.
[7] A. A. Aquino-Lugo, R. Klump, and T. J. Overbye, A control framework
for the smart grid for voltage support using agent-based technologies,
IEEE Trans. Smart Grid, vol. 2, no. 1, pp. 161168, Mar. 2011.
[8] E. M. Lightner and S. E. Widergren, An orderly transition to a
transformed electricity system, IEEE Trans. Smart Grid, vol. 1, no. 1,
pp. 310, Jun. 2010.
[9] G. Zhabelova and V. Vyatkin, Multiagent smart grid automation archi-
tecture based on IEC 61850/61499 intelligent logical nodes, IEEE Trans.
Ind. Electron., vol. 59, no. 5, pp. 23512362, May 2012.
[10] S. Bruno, S. Lamonaca, G. Rotondo, U. Stecchi, and M. La Scala, Un-
balanced three-phase optimal power flow for smart grids, IEEE Trans.
Ind. Electron., vol. 58, no. 10, pp. 45044513, Oct. 2011.
[11] P. Tenti, A. Costabeber, P. Mattavelli, and D. Trombetti, Distribution
loss minimization by token ring control of power electronic interfaces
in residential microgrids, IEEE Trans. Ind. Electron., vol. 59, no. 10,
pp. 38173826, Oct. 2012.
[12] M. A. Zamani, T. Sidhu, and A. Yazdani, Investigations into the control
and protection of an existing distribution network to operate as a micro-
grid: A case study, IEEE Trans. Ind. Electron., vol. 61, no. 4, pp. 1904
1915, Apr. 2014.
[13] A. Vaccaro, G. Velotto, and A. F. Zobaa, A decentralized and cooperative
architecture for optimal voltage regulation in smart grids, IEEE Trans.
Ind. Electron., vol. 58, no. 10, pp. 45934602, Oct. 2011.
[14] V. Calderaro, C. N. Hadjicostis, A. Piccolo, and P. Siano, Failure iden-
tification in smart grids based on petri net modeling, IEEE Trans. Ind.
Electron., vol. 58, no. 10, pp. 46134623, Oct. 2011.
[15] M. Sechilariu, B. Wang, and F. Locment, Building integrated photo-
voltaic system with energy storage and smart grid communication, IEEE
Trans. Ind. Electron., vol. 60, no. 4, pp. 16071618, Apr. 2013.
[16] V. C. Gungor, L. Bin, and G. P. Hancke, Opportunities and challenges
of wireless sensor networks in smart grid, IEEE Trans. Ind. Electron.,
vol. 57, no. 10, pp. 35573564, Oct. 2010.
[17] T. Sauter and M. Lobashov, End-to-End communication architecture for
smart grids, IEEE Trans. Ind. Electron., vol. 58, no. 4, pp. 12181228,
Apr. 2011.
[18] N. Liu, J. Chen, L. Zhu, J. Zhang, and Y. He, A key management
scheme for secure communications of advanced metering infrastructure in
smart grid, IEEE Trans. Ind. Electron., vol. 60, no. 10, pp. 47464756,
Oct. 2013.
[19] R. Smolenski, J. Bojarski, A. Kempski, and P. Lezynski, Time-domain-
based assessment of data transmission error probability in smart grids with
electromagnetic interference, IEEE Trans. Ind. Electron., vol. 61, no. 4,
pp. 18821890, Apr. 2014.
[20] H. Xin, Z. Qu, J. Seuss, and A. Maknouninejad, A self-organizing
strategy for power flow control of photovoltaic generators in a distribu-
tion network, IEEE Trans. Power Syst., vol. 26, no. 3, pp. 14621473,
Aug. 2011.
[21] M. Shahraeini, M. H. Javidi, and M. S. Ghazizadeh, Comparison between
communication infrastructures of centralized and decentralized wide
area measurement systems, IEEE Trans. Smart Grid, vol. 2, no. 1,
pp. 206211, Mar. 2011.
[22] S. Bou Ghosh et al., Agent-oriented designs for a self healing smart grid,
in Proc. SmartGridComm Conf., 2010, pp. 461466.
[23] NIST framework and roadmap for smart grid interoperability stan-
dards, Release 1.0, NIST, Gaithersburg, MD, USA, Jan. 2010. [Online].
Available: http://www.nist.gov/public_affairs/
[24] IEEE Recommended Practice for Interconnecting Distributed Resources
with Electric Power Systems Distribution Secondary Networks, IEEE Std.
1547.6-2011, Sep. 12, 2011.
[25] M. Kezunovic, Smart fault location for smart grids, IEEE Trans. Smart
Grid, vol. 2, no. 1, pp. 1122, Mar. 2011.
[26] S. A. Arefifar, Y. A.-R. I. Mohamed, and T. El-Fouly, Comprehensive
operational planning framework for self-healing control actions in smart
distribution grids, IEEE Trans. Power Syst., vol. 28, no. 4, pp. 4192
4200, Nov. 2013.
[27] M. Shahidehpour and Y. Wang, Communication and Control in Electric
Power Systems. Hoboken, NJ, USA: Wiley, 2003.
[28] B. Yang, V. Vittal, and G. T. Heydt, Slow-coherency-based controlled
islandingA demonstration of the approach on the August 14, 2003
blackout scenario, IEEE Trans. Power Syst., vol. 21, no. 4, pp. 1840
1847, Nov. 2006.
[29] G. Xu and V. Vittal, Slow coherency based cutset determination algo-
rithm for large power systems, IEEE Trans. Power Syst., vol. 25, no. 2,
pp. 877884, May 2010.
[30] J. Li, C.-C. Liu, and K. P. Schneider, Controlled partitioning of a power
network considering real and reactive power balance, IEEE Trans. Smart
Grid, vol. 1, no. 3, pp. 261269, Dec. 2010.
[31] Y. Attwa, E. F. El-Saadany, M. M. A. Salama, and R. Seethapathy,
Optimal renewable resources mix for distribution system energy loss
minimization, IEEE Trans. Power Syst., vol. 25, no. 1, pp. 360370,
Feb. 2010.
[32] J. M. S. Pinheiro, C. R. R. Dornellas, M. Th. Schilling, A. C. G. Melo, and
J. C. O. Mello, Probing the new IEEE Reliability Test System (RTS-96):
HL-II assessment, IEEE Trans. Power Syst., vol. 13, no. 1, pp. 171176,
Feb. 1998.
[33] S. A. Arefifar and Y. A.-R. I. Mohamed, DG mix, reactive sources
and energy storage units for optimizing microgrid reliability and supply
security, IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 18351844, Jul. 2014.
[34] J. S. Savier and D. Das, Impact of network reconfiguration on loss
allocation of radial distribution systems, IEEE Trans. Power Del., vol. 22,
no. 4, pp. 24732480, Oct. 2007.
[35] R. Kim, J. U. Lee, and J.-B. Jo, Hierarchical spanning tree network
design with Nash genetic algorithm, Comput. Ind. Eng., vol. 56, no. 3,
pp. 10401052, Apr. 2009.
[36] B. V. Cherkassky, A. V. Goldberg, and T. Radzik, Shortest paths algo-
rithms: Theory and experimental evaluation, Math. Programm. Ser. A,
vol. 73, no. 2, pp. 129174, May 1996.
[37] M. E. Baran and F. F. Wu, Optimal capacitor placement on radial dis-
tribution systems, IEEE Trans. Power Del., vol. 4, no. 1, pp. 725734,
Jan. 1989.
[38] M. E. H. Golshan and S. A. Arefifar, Distributed generation, reactive
sources and network-configuration planning for power and energy-loss
reduction, Proc. Inst. Elect. Eng.Gener., Transmiss. Distrib., vol. 153,
no. 2, pp. 127136, Mar. 2006.
Seyed Ali Arefar (S06M10) was born in
Isfahan, Iran. He received the B.Sc. and M.Sc.
degrees (with honors) in electrical engineering,
power systems, from Isfahan University of Tech-
nology, Isfahan, in 2001 and 2004, respectively,
and the Ph.D. degree in electrical engineering
from the University of Alberta, Edmonton, AB,
Canada, in 2010.
From 2011 to 2014, he was a Natural Sci-
ences and Engineering Research Council of
Canada Visiting Fellow with CanmetENERGY,
Natural Resources Canada (NRCan), and the visiting Postdoctoral
Program with the University of Alberta. He is currently with the University
of Alberta as a Research Associate. His research interests include
optimization in planning and operation of smart grids, load modeling,
and power quality.
AREFIFAR et al.: CLUSTERING OF SYSTEMS CONSIDERING COMMUNICATION AND CONTROL REQUIREMENTS
Yasser Abdel-Rady I. Mohamed (M06SM11)
was born in Cairo, Egypt, on November 25,
1977. He received the B.Sc. (with honors) and
M.Sc. degrees in electrical engineering from
Ain Shams University, Cairo, Egypt, in 2000
and 2004, respectively, and the Ph.D. degree
in electrical engineering from the University of
Waterloo, Waterloo, ON, Canada, in 2008.
He is currently with the Department of Elec-
trical and Computer Engineering, University of
Alberta, Edmonton, AB, Canada, as an Asso-
ciate Professor. His research interests include dynamics and control
of power converters; distributed and renewable generation, and micro-
grids; modeling, analysis, and control of smart grids; electric machines;
and motor drives.
Dr. Mohamed is an Associate Editor of the IEEE T RANSACTIONS
ON I NDUSTRIAL E LECTRONICS . He is also a Guest Editor of the IEEE
T RANSACTIONS ON I NDUSTRIAL E LECTRONICS Special Section on Dis-
tributed Generation and Microgrids. His biography is listed in Marques
Whos Who in the World. He is a Registered Professional Engineer in
the Province of Alberta, Canada.
723
Tarek El-Fouly received the B.Sc. and
M.Sc. degrees in electrical engineering from
Ain Shams University, Cairo, Egypt, in 1996
and 2002, respectively, and the Ph.D. degree
in electrical engineering from the University of
Waterloo, Waterloo, ON, Canada, in 2008.
He joined CanmetENERGY, Natural Re-
sources Canada, Varennes, QC, Canada, in
2008, as a Transmission and Distribution Re-
search Engineer, where he is conducting and
managing research activities related to active
distribution networks, microgrids, and remote communities. In 2010,
he was appointed as Adjunct Assistant Professor with the Department
of Electrical and Computer Engineering, University of Waterloo. His
research interests include protection and coordination studies, integra-
tion of renewable energy resources, smart microgrid, smart remote
community applications, demand-side management, and forecasting.
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014 209

Short-Term Operation Scheduling in

Renewable-Powered Microgrids:
A Duality-Based Approach
Binyan Zhao, Student Member, IEEE, Yi Shi, Member, IEEE, Xiaodai Dong, Senior Member, IEEE,
Wenpeng Luan, Senior Member, IEEE, and Jens Bornemann, Fellow, IEEE

AbstractThis paper considers the operation scheduling prob- UC indication function.

lem in renewable-powered microgrids, which is used to determine Emission limit.
the least-cost unit commitment (UC) and the associated dispatch, Euclidean projection of to the interval
while meeting load, environmental, and system operating require-
ments. The intermittency nature of the renewable energy sources, , i.e., .
as well as microgrids capacity to operate either in parallel with, or Euclidean projection of to the interval
autonomously of, the traditional power grid, pose new challenges to , i.e., .
this classic optimization task. A probability-based concept, proba- Fuel cost coefcients of a unit.
bility of self-sufciency (PSS), is introduced to indicate the proba- Aggregated cost coefcients of a unit.
bility that the microgrid is capable of meeting local demand in a
self-sufcient manner. Furthermore, to the best of our knowledge, Maintenance cost coefcients of a unit.
we make the rst attempt in approaching the mixed-integer UC Error function [17, eq. 8.250.1].
problem from a convex optimization perspective, which leads to an Unit index (subscript).
analytical closed-form characterization of the optimal commitment
Iteration index (superscript).
and dispatch solutions. The simulation results show that 1) the
proposed method achieves an efcient performance that incurs no Power dispatch of a unit.
loss of optimality with lower complexity than existing algorithms; Optimal power dispatch of a unit.
2) an energy storage system (ESS) with suitable capacity contributes , Minimum and maximum power generation
to the self-sufciency target of a microgrid, and the stored energy
limits.
varies less remarkably as the microgrid tends to operate more
independently; 3) the proposed method provides guidelines in decid- Wind turbine power output.
ing the ESS size to achieve a desired PSS. Power dispatch of a unit in the relaxed UC
Index TermsDistributed generators, duality, economic dispatch, problem.
microgrid, renewable energy resources, unit commitment (UC). Forecasted wind power in time interval .
Hour index (subscript).
NOMENCLATURE Unit status indicator of a unit where 1 means
ON and 0 means OFF.
Symbol Description
Cut-in wind speed.
Emission coefcients of a unit.
Cut-out wind speed.
Demand forecast error.
Rated wind speed.
Wind power forecast error.
Wind speed.
Convergence criterion parameter.
Rated electrical power.
, , Lagrange multipliers.
Modiedelectricitydemandintimeinterval .
Mean of demand forecast error.
, Lower and upper limits of the energy stored
Mean of wind power forecast error.
in ESS.
Cooling time constant of a unit.
Cold start-up cost of a unit.
Variance of demand forecast error. Dual function.
Variance of wind power forecast error. Forecasted demand in time interval .
Step size of the subgradient method. Emission function of a unit.
CDF of wind power forecast error.
Manuscript received November 23, 2012; revised April 12, 2013 and July 15, Fuel cost function of a unit.
2013; accepted July 16, 2013. Date of publication September 30, 2013; date of
current version December 12, 2013. Hot start-up cost of a unit.
B. Zhao, Y. Shi, X. Dong, and J. Borenmann are with the Department of L Lagrange function.
Electrical and Computer Engineering, University of Victoria, Victoria, BC V8W Maintenance cost function of a unit.
3P6, Canada (e-mail: binyanz@ece.uvic.ca; shiyipfc@ece.uvic.ca; xdong@ece.
uvic.ca; jbornema@ece.uvic.ca). , The time a unit needs to remain ON/OFF if ON/
W. Luan is with the State Grid Smart Grid Research Institute, Beijing 100085, OFF at the beginning of the scheduling period.
China (e-mail: luanwenpeng@sgri.sgcc.com.cn). Total number of units.
Color versions of one or more of the gures in this paper are available online at
http://ieeexplore.ieee.org. Probability operator.
Digital Object Identier 10.1109/TSTE.2013.2279837 Probability of self-sufciency.

1949-3029 2013 IEEE

210
Operating reserve requirement in time
interval .
Start-up cost function of a unit.
Total time horizon of 1 day.
Total cost of all units.
, The time a unit has been ON/OFF at the
beginning of the scheduling period.
I. INTRODUCTION
USTAINABILITY has become an imperative requirement
S on the electric power infrastructure with the impending
energy crisis and environment deterioration. In parallel to the
rapid evolution of smart-grid-based solutions [1], another emerg-
ing power distribution system known as microgrids is also
quietly gaining momentum. A microgrid is an integrated system
consisting of a set of distributed generators (DGs) [microturbines
(MT), fuel cells (FC), reciprocating engines] and renewable
energy sources (RESs) [solar photovoltaics (PV) and wind
turbines (WT) systems] [2], which function cooperatively to
furnish the cooperation in parallel with, or autonomously of, the
traditional electricity macrogrid. The dening characteristic of
a microgrid is its ability to separate itself seamlessly from the
macrogrid during a utility grid disturbance, and function as a
self-controlled entity with high efciency and low greenhouse
gas emissions. This transparent adaptation in operational modes,
along with the capacity to better manage distributed energy
resources (DERs), renders microgrid the most promising solu-
tion to develop a more reliable and decentralized energy system.
The daily operation of a microgrid involves nding the least-
cost dispatch of the DGs that minimizes the total operating cost,
while meeting the electrical load and satisfying various technical,
environmental, and operating constraints. This can be seen as a
downsized version of the unit commitment (UC) problem that is
traditionally applied to large central generators in macrogrid [3].
Mathematically, the UC task can be described as a mixed-integer
optimization problem with a nonlinear solution space, which has
been the subject of intensive investigation for more than 40 years
(c.f., bibliographical survey [4]). Nevertheless, the unique features
of microgrids introduce further restrictions as well as simplica-
tions to this classic optimization task that needs to be addressed.
The biggest challenge comes from the intermittent nature of
RESs, which often leads to power variations and makes it much
more difcult to produce accurate day-ahead schedules in
microgrids. Therefore, the operation scheduling of the dispatch-
able DGs should be performed at a much ner level, providing
quick and continuous power provision, based on the most recent
and accurate forecasted data [5]. Although large thermal units
are often subject to ramping rate limits that are typically in the
order of several tens of megawatts per hour, it only takes
several minutes for an MT to ramp up from 0 to full load [6].
This further supports the operation scheduling to be performed
at a more granular level for microgrids [7]. Last but not the least,
the capacity of switching between different operational modes
calls for a proper modeling framework that reects the unique
characteristic of microgrids.
Various numerical-based algorithms, including some typical
heuristic methods such as genetic algorithm (GA) [8], [9],
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014
particle swarm optimization (PSO) [10], simulated annealing
techniques [11], [12], and network-ow programming [13], have
been proposed to solve the optimal UC problem, but most of
them only provide a reasonable numerical solution (suboptimal,
nearly global optimal) and have high computational complexity.
For the classical mixed-integer UC problem, the branch and
bound (BB) technique provides exact accurate numerical results.
However, the procedure is generally not efcient except when
large portions of the solution space can be quickly discarded in
the case that there are not too many solutions having near optimal
function values. In [14], a linear optimization problem was
formulated and numerically solved to determine the optimal
size of energy storage system (ESS) that minimizes the operating
cost. The authors in [15] and [7] also established UC strategies
based on piecewise linear blocks approximating the quadratic
objective function and frequency droop scheme. The authors in
[16] modeled explicitly the length of time the microgrid operates
autonomously and used a Monte Carlo analysis to study the
impact of RESs over a set of United Kingdom commercial
load proles.
This paper deals with a quadratic formulation of the environ-
mental/economic UC optimization problem in a microgrid that
consists of DGs, RESs, and an ESS. The contributions of this
work are summarized as follows: 1) in light of microgrids
unique operational feature and to take into account the forecast
errors that exist in demand and wind power forecast, a novel
probability-based concept is proposed to indicate the probability
that the microgrid is able to operate in islanded mode, termed
probability of self-sufciency (PSS); 2) the mixed-integer UC
problem is approached from a convex optimization perspective,
which leads to an analytical closed-form solution. Compared to
two classical methods, BB and GA, the proposed method uses
signicantly less processing time yet renders no loss of optimal-
ity in performance; and 3) the proposed method provides guide-
lines in deciding the size of ESS to improve the autonomous
target PSS for a microgrid efciently.
The remainder of this paper is organized as follows. Section II
discusses the basic system models including wind power forecast,
operating cost, emission, as well as various unit and operation
constraints. In Section III, the optimal UC problem is formulated,
transformed into a convex optimization problem, and nally
solved in analytical closed forms. An efcient subgradient-based
algorithm is also proposed to obtain the optimal UC solution in
this section. Section IV conducts numerical simulations that
verify the efciency and reliability of the proposed scheme.
Finally, concluding remarks are drawn in Section V.
II. SYSTEM MODEL
We consider a microgrid that consists of a set of DG units,
including both MTs and FCs, an RES, the WT, and an energy
storage module, the ESS. Electrical loads in the microgrid are
prioritized into tiers, which consists of critical loads that relate to
essential processes that must be always met and lower-priority
non-critical loads that can be temporarily removed until adequate
power is available. Due to the limited capacity of the DERs and
as in [18][20], we assume that the microgrid schedules its units
to meet the critical loads in the highest priority, satises the
ZHAO et al.: SHORT-TERM OPERATION SCHEDULING IN RENEWABLE-POWERED MICROGRIDS
non-critical loads using best efforts, and purchases power from
the macrogrid in case of supply shortage. In addition, to better
utilize an environmentally friendly resource, we assume that the
WT is always ON and functions as the primary power source [14].
The DGs, on the other hand, serve as backup generators and work
collaboratively with the WT to meet the critical loads. The ESS
module is introduced to mitigate the renewable power intermit-
tencies and load mismatches. In this work, we assume that the
microgrid updates its UC strategy every 1 h1 during which load
and generation are considered constant.
A. Forecasted Wind Power Model
Wind power is the electrical power generated by WT, installed
in locations with strong and sustained winds. In practice, the
actual wind power almost entirely depends on the wind speed
when other physical limitations are xed or changed relatively
slowly [14]. In principle, is a random variable and varies
continuously over time. In this work, we assume that remains
unchanged in one scheduling period (i.e., 1 h), and can vary
independently between different scheduling periods.
Extensive research has been done in developing wind
forecasting models and approaches [14], [21]. To capture the
relationship between the wind speed and wind power and as in
[14] and [22], the following piecewise linear model is adopted:
< function is typically expressed as a polynomial, the order of
B. Cost Models for MT and FC
MTs are small electricity generators that burn gaseous and
liquid fuels to create high-speed rotation that turns an electrical
generator. Depending on the size range, an MT can ramp up
from 0 to full load between 10 s to several minutes [6]. The FC
technology uses an electrochemical process rather than a com-
bustion process to generate electricity. Polymer electrolyte FCs,
also known as proton exchange membrane FCs, are particularly
attractive for microgrids that require rapid start-up and quick
response to load change [23].
The operating cost of an MT/FC usually includes fuel cost,
maintenance cost, and start-up cost.
Fuel Costs: The fuel costs for DGs are considered as a
quadratic model [24], which includes linear fuel cost models
as special cases [7], [14]. The fuel costs of unit in time interval
can be expressed as
Maintenance Costs: The maintenance costs for DGs are based
on forecasts with minimal real-life situations, which are assumed
to be proportional to the produced power [25]. Therefore, the
maintenance costs of unit in time interval is
1 2
Depending on different application requirements, a smaller updating interval
can be selected since the closed-form UC solutions can be computed fast.
211
Start-Up Costs: The generator start-up cost depends on the
time the unit has been off prior to a start-up. The start-up cost
of unit in time interval can be represented by an exponential
cost curve [24, eq. (3.12)]
Sometimes, the industry is interested in the total cost per day
C. Emission Model
Emission effects should be taken into account for an
environmentally friendly power production. The microgrids are
envisioned to be new energy savings and green grids in the
future, which entails carbon emissions limited to regulations and
legal requirements.
The amount of emissions produced depends on the fuel used,
pollution control devices installed, and the amount of electricity
generated. In this work, we assume that only DGs produce
emissions and the RESs are emission free.2 The emission
which depends on the desired accuracy [27], [28]. As in [15] and
[28], a quadratic function is considered for the emission curve
as follows:
D. Energy Storage System
In renewable-powered microgrids, the problem of mitigating
power intermittencies and load mismatches is an important and
challenging task. In this context, an ESS plays a critical role in
shaving peak demand and compensating forecast errors. For
instance, when the forecasted wind power is smaller than the
actual value (i.e., an underestimate), the supplied power is likely
to be larger than the actual electricity demand, in which case
the ESS will be functioning in charging state to store surplus
electrical/renewable energy, which can be dispatched properly
later in the event of power shortage.
The charge and discharge of the ESS is subject to stored energy
limits, and , which specify the minimum and maxi-
mum energy stored in the battery bank, respectively. In this case,
is set as the full capacity of the ESS and to be around
10% of its full capacity. The ESS is also subject to starting and
ending limits that specify the initial and nal energy inside the
battery bank during the course of 1 day. In this work, the starting
and ending limits are both selected as for the purpose of
energy balance of the ESS [14].
Note that the emissions in the production process of WT and other equipments
[26] have not been considered.
212 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014

E. Unit and Operation Constraints As in [29] and [30], we assume that both demand forecast
error and wind power forecast error can be modeled
Maximum and Minimum Output Limits: The output power of
as independent normally distributed random variables, i.e.,
the DG in stable operation is restricted by its lower and upper
and , respectively. Then,
limits as follows:
the probabilistic power balance constraint can be expressed as

In our case, the minimum available power for the DGs is 0, i.e.,
the DG can be turned OFF when the output power from the WT is
enough to meet the demanded power. which, after some algebra, can be reformulated as
Minimum Up/Down Time: Once a DG is switched ON, it has to
operate continuously for a certain number of times before it can
be switched OFF. Also, a certain number of hours have to pass
before a DG can be brought online after being switched OFF.
Violation of such constraints can shorten the units life time.
Mathematically, we have

III. PROBLEM FORMULATION AND CLOSED-FORM SOLUTIONS

For DGs in microgrids, the minimum up/down time of DGs is
around 600 and 300 s [5], respectively, which is always satised
under hourly scheduling operations.
Ramp Rates: Traditional thermal units are often subject to ramp
rate limits that specify the amount a units generation can increase
or decrease during one scheduling period. In the context of
microgrids, small DG units can ramp up from 0 to full load in
the order of several minutes [23]. Thus, the ramp rate limits are
typically not reached under normal hourly scheduling operations.
Emission Limits: To comply with the purpose of environment
conservation and reduce the greenhouse gas footprint, we impose
hourly emission limits on all the DGs. Mathematically, we have
Based on the aforementioned cost models and system con-
straints, we rst formulate the UC optimization problem that is
used to determine the loads and commitments of DGs such that
the total cost is minimized, power demand is met, and the system
constraints are satised. A duality-based analysis is then applied
to derive closed-form solutions. Finally, an efcient subgradient-
based method is introduced to numerically compute the optimal
solutions.
A. Problem Formulation
The UC optimization problem in a particular operation sched-
uling interval can be written as

Operating Reserves: In the event of a power supply disruption,

Evidently, problem is a mixed-integer programming prob-
operating reserve constraints guarantee that there exist extra
lem with a nonlinear solution space. In order to transform into
generating capacity to the system that can be brought online
a convex optimization problem, we introduce auxiliary power
immediately (spinning reserves) or within a short interval
variables and relax to be a continuous variable
(supplementary reserves). In microgrids with fast-start DGs,
in [0,1]. Thus, the transformed problem can be written as3
operating reserves are imposed as follows [3]:

Probability of Self-Sufciency: Both demand and renewable

power forecast are prone to errors, which negatively affect
the microgrid in meeting the local power demand and their
autonomous and independent functions. Once a microgrid
cannot meet the power demand solely based on local
generating units, it can switch to a grid-connected mode and
purchase energy from the upstream macrogrid. To better
understand the impacts of operational mode on total operating
cost, we propose the use of a novel probability-based concept,
PSS, which indicates the target probability that the microgrid
is able to operate in islanded mode without purchasing energy 3
The objective function is dened at by continuity as
from the macrogrid. .
ZHAO et al.: SHORT-TERM OPERATION SCHEDULING IN RENEWABLE-POWERED MICROGRIDS 213

where the aggregated cost coefcients are and Besides, we observe that the commitment decisions are
. Note that the objective function of completely determined by the sign of at the optimal
is equivalent to that of as the change of variables is invertible dispatch [c.f. (18)]. Thus, can be seen as a UC
except for . The same statement holds for the emission indication function. In the case that the power generated by
constraints (10) and (14). In case that , every the WT is sufcient to meet local power demand, the microgrid
solves . will be entirely powered by the RES and no DG needs to be
The Hessian matrices of the objective function and the turned ON, i.e., for all .
constraints of are positive semi-denite, and the form of
the whole problem meets the requirement of a convex problem C. Subgradient-Based Algorithm
[31], indicating that is a convex problem. So far, we have found the analytical optimal solutions in
(17) and in (18), which are, however, functions of the
B. Closed-Form Economic Dispatch and UC Solutions
Lagrangian multipliers. By convexity, it sufces to obtain the
To solve analytically, we rst decouple the optimization optimal dual variables ( , , ) to the dual problem, which
variables and by substituting . Note that are used to compute optimal primal solutions (17) and (18).
for the case of interest, i.e., . For this reason, The complete algorithm is formally stated in Algorithm 1.
we will henceforth use . The Lagrangian function can This section discusses the use of a subgradient-based iterative
then be written as procedure to numerically compute the corresponding optimal
Lagrangian multipliers , , and . The key iteration steps
L are [31]

where can be viewed as the UC indication function

The dual function can be obtained by minimizing the which are provably convergent to the optimal value provided
Lagrangian (14), which is given by that the step sizes are selected to satisfy and
< [32]. A graphical convergence illustration can
be found in the simulation section.

Algorithm 1 Subgradient-based algorithm

Initialize Lagrangian multiplier , , and to arbitrary

non-negative values and set the step size .
The initial step to solve (16) is to tackle the inner minimization,
the solution of which can be easily derived as repeat
Compute economic dispatch using (17).
Obtain UC indication function via (15).
Find unit status indicator using (18).
Update , , and via (19), (20), and (21),
The next step is to solve the outer minimization over in respectively.
(16), given the optimal power dispatch solution (17). Evidently,
the minimum value is attained by setting for all until < .
< and otherwise. Mathematically, we have

< IV. SIMULATION RESULTS AND DISCUSSION

In this study, a microgrid system consisting of two MTs, FC
and WT, and one ESS is considered for a scheduling time horizon
Remark 1: An important step in the course of analytical of 24 h. The fuel cost coefcients, the emission coefcients,
derivation is to decouple the optimizations of the commitment and the power limits of the DGs are summarized in Table I, where
status and the power dispatch separately as in (16). of all DGs are 0 kW. Similar parameter settings have also
214
TABLE I
PARAMETERS OF DISTRIBUTED GENERATORS
Fig. 1. Total operating cost per day under different PSS targets.
been used in [8]. The emission limit is set as 150 kg/hr.
The wind speed data samples adopted in this work are from
the Wind Test Center in West Texas A&M University [33],
with the parameters in (1),
. Unless stated otherwise, the demand forecast
error statistics and those of the wind power are set as
, , and . To understand the
impact of PSS on the total operating cost, three different
operation scenarios are considered, which are , processor at 2.67 GHz.
, and .
A. UC, Dispatching and Methods Comparison
This section illustrates the derived UC solutions without
considering the functionality of the ESS. The effects of ESS
will be investigated in Section IV-C. Fig. 1 illustrates the fast
convergence behavior of the proposed algorithm in optimizing
the total operating cost per day of the microgrid. It takes only
about 20 iterations to reach the optimal solution under all three
PSS values. Besides, it is observed from Fig. 1 that the operating
cost increases as the microgrid functions more autonomously,
since more power has to be generated to ensure self-sufciency
and to mitigate demand and wind power forecast errors.
Fig. 2 shows the comparison results of four different methods
in our environmental/economic dispatch optimization problem.
We use GA and the proposed method (Prop-algor) to solve the
original problem and the transformed problem , respec-
tively. In addition, BB is applied to solve , providing us an
accurate result as a benchmark. In terms of computational
complexity, the proposed method has [31] time
complexity with the number of generators in the system,
while BB with depth-rst search requires the computation on the
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014
Fig. 2. Performance comparison of different optimization algorithms under
. , , and
for GA.
Fig. 3. Elapsed CPU time of the proposed algorithm under different stopping
criteria. The computer used was an ThinkPad Laptop with a i7 M560 duo-core
order of , with < < [34], and a simple GA
involves a complexity of , where gens is the
number of generations and pop the population size [35]. There-
fore, the proposed method achieves accurate results with lower
computational complexity compared to BB and GA as shown in
Fig. 2. The corresponding CPU time to run the proposed algo-
rithm has been plotted in Fig. 3. It is observed that the algorithm
running time decreases as the stopping criterion drops from
within 0.1% of the value in the previous iteration to 10%.
The optimal amount of dispatch of each DG for given demand
and wind power forecast proles is shown in Fig. 4 under
different target PSS values. Take Fig. 4(b) as an example.
Observe the forecast wind power decreases from 1:00 A.M. to
5:00 A.M. in the early morning. In order to fulll the demand, the
DGs are dispatched economically according to the optimal UC
solution. The dispatched power from the DGs continues to be in
a higher level until 13:00 P.M., when the forecast wind power
comes into play again. At midnight, the power demand reaches a
minimum and the microgrid is entirely powered by the wind
power. During the whole process, the DGs serve as backup
ZHAO et al.: SHORT-TERM OPERATION SCHEDULING IN RENEWABLE-POWERED MICROGRIDS 215

Fig. 4. Forecasted demand and wind power as well as optimal dispatch of the DGs under different PSS targets: (a) , (b) , and
.

TABLE II
UNIT COMMITMENT OF THE DISTRIBUTED GENERATORS WITH TARGET

sources that complement the renewable source in meeting elec-

tricity demand.
Table II further depicts the UC status of the DGs for the case of
over a period of 24 h. Cross-referencing Fig. 4(b),
we can observe that an FC is the most preferred power source
among the DGs and contributes signicantly during the entire
scheduling period. In contrast, MT2 contributes the least and Fig. 5. Total operating cost per day under different time horizons.
is always the last one to be turned ON. The reason is as follows.
Recall from (18) in Section III-B that the commitment status of
the DG is solely determined by the sign of the UC indication
function at the optimal dispatched point . An interest-
ing phenomenon observed from numerical results is that
< < for all . In other
words, there exists a hidden priority listing order that arranges
the DGs based on lowest operational cost characteristics as well
as emissions.

B. Operating Cost Versus Different Forecasting Time Horizons

Clearly, the shorter the forecasting interval is, the more
accurate the forecast data and the smaller the variance of the
forecast error. According to [29] and [36], the typical standard
deviation of the wind power forecast error for a specic wind
farm can be expressed as a function of the forecast horizon, which Fig. 6. Achieved PSS versus the capacity of the ESS based on wind power
can be approximated accurately by a linear function when the statistics collected in 1 month.
forecast horizon is less than 6 h. Fig. 5 examines the impact of
varied forecasting time horizons on the total operating cost,
C. The Impact of ESS on Microgrids Autonomy
based on the aforementioned linear model. It is observed that for
the case of , the operating cost grows almost linearly In the last numerical example, we incorporate the ESS into the
as the forecasting interval increases from 10 min to 1 h. The microgrid setting and study the impact of ESS on the achieved
same trend is observed for the case of . Interestingly, level of autonomy of the microgrid. To verify the effectiveness of
the cost holds constant for the case of , due to the fact the ESS on the achieved PSS, data samples collected by the wind
that the microgrid is indifferent to either being connected to, or test center [33] in 1 month are employed to calculate the achieved
autonomous of the macrogrid. PSS of the microgrid in practice. We observe from Fig. 6 that for
216
Fig. 7. Variations in the amount of energy stored in ESS in 1 day under different PSS targets: (a)
all the PSS targets, 1) the proposed approach successfully meets
the design targets in the absence of ESS. For example, the
microgrid achieves a practical PSS of 52%, 78%, and 92% when
the PSS targets are set as , , and , respectively; 2) the
microgrid is more capable of functioning self-sufciently as the
capacity of ESS increases [37]; and 3) the achieved PSS hits an
upper limit when the capacity of ESS is larger than a threshold.
From our simulations, this ESS threshold grows from 40 kWh, to
90 kWh, and to 120 kWh as the PSS targets decreases from 90%
to 70%, and to 50%, respectively. This provides a guideline on
determining the ESS size to achieve a desired PSS level.
To further analyze the variations of the energy stored in the
ESS in 1 day, we plot in Fig. 7 the evolution of the energy
stored in ESS under different target PSSs. The capacity of ESS
is set as 60, 90, and 120 kWh as in Fig. 7(a), (b), and (c) when
the PSS targets are equal to 90%, 70%, and 50%, respectively.
Due to practical concerns, the starting ( ) and ending limits
( ) of ESS in this work are set as around 10% of the
capacity of ESS, which are 10 kWh for all the cases. We
then observe that the events of charging and discharging
occur frequently during the time period between 0:00 A.M. and
10:00 A.M., when the forecasted wind power falls short of the
forecasted demand. Also note that the stored energy in the ESS
varies less remarkably as the microgrid tends to operate more
independently.
V. CONCLUSION
The unique characteristics of renewable-powered microgrid
have brought new challenges to the classic UC optimization task
of UC. We have shown that the traditional problem formulation
can be modied to incorporate the intermittency of the RESs,
emission limits on the carbon footprint, as well as the forecast
error that exists in demand and renewable power forecasts. Using
a duality-based approach, it has been demonstrated that an
analytical characterization of the optimal commitment and dis-
patch solutions for the DGs is available, which can be computed
very efciently using a subgradient-based algorithm. The ap-
proach can be easily modied to incorporate other types of DGs
or RESs. Our study shows that the features of DGs can have a
great impact on the operation of UC strategy in microgrids,
which will be investigated in our future work.
IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014
, (b) , and (c) .
REFERENCES
[1] V. C. Gngr, D. Sahin, T. Kocak, S. Ergt, C. Buccella, C. Cecati, and
G. P. Hancke, Smart grid technologies: Communications technologies and
standards, IEEE Trans. Ind. Inform., vol. 7, no. 4, pp. 529539, Nov. 2011.
[2] A. K. Basu, S. Chowdhury, and S. Paul, Microgrids: Energy management
by strategic deployment of DERsA comprehensive survey, Renew.
Sustain. Energy Rev., vol. 15, no. 9, pp. 43484356, Dec. 2011.
[3] A. Cohen and V. Sherkat, Optimization-based methods for operations
scheduling, Proc. IEEE, vol. 75, no. 12, pp. 15741591, Dec. 1987.
[4] N. Padhy, Unit commitmentA bibliographical survey, IEEE Trans.
Power Syst., vol. 19, no. 2, pp. 11961205, May 2004.
[5] S. A. Pourmousavi, M. H. Nehrir, C. M. Colson, and C. Wang, Real-time
energy management of a stand-alone hybrid wind-microturbine energy
system using particle swarm optimization, IEEE Trans. Sustain. Energy,
vol. 1, no. 3, pp. 193201, Oct. 2011.
[6] R. Yinger, Behavior of Capstone and Honeywell microturbine generators
during load changes, California Energy Commission, Tech. Rep. P500-
04-021, 2001.
[7] C. Hernandez-Aramburo, T. Green, and N. Mugniot, Fuel consumption
minimization of a microgrid, IEEE Trans. Ind. Appl., vol. 41, no. 3,
pp. 673681, May 2005.
[8] C. Chen, S. Duan, T. Cai, B. Liu, and G. Hu, Smart energy management
system for optimal microgrid economic operation, IET Renew. Power
Gen., vol. 5, no. 3, pp. 258267, May 2011.
[9] T. Logenthiran and D. Srinivasan, Short term generation scheduling of a
microgrid, in Proc. IEEE TENCON, Singapore, Nov. 2009, pp. 16.
[10] M. R. Alrashidi and M. E. El-Hawary, Impact of loading conditions on
the emission-economic dispatch, World Academy of Science, Eng. and
Technol., vol. 15, no. 29, pp. 148151, Mar. 2008.
[11] R. Brown, S. Gupta, R. Christie, S. Venkata, and R. Fletcher, Automated
primary distribution system design: Reliability and cost optimization,
in Proc. IEEE Transmission and Distribution Conf., Los Angeles, CA,
USA, Nov. 1996.
[12] S. Jonnavithula and R. Billinton, Minimum cost analysis of feeder routing
in distribution system planning, IEEE Trans. Power Del., vol. 11, no. 4,
pp. 19351940, Oct. 1996.
[13] Y. Tang, Power distribution system planning with reliability modeling and
optimization, IEEE Trans. Power Syst., vol. 11, no. 1, pp. 181189,
Feb. 1996.
[14] S. Chen, H. Gooi, and M. Wang, Sizing of energy storage for microgrids,
IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 142151, Mar. 2012.
[15] A. Abdollahi, M. P. Moghaddam, M. Rashidinejad, and M. K. Sheikh-El-
Eslami, Investigation of economic and environmental-driven demand
response measures incorporating UC, IEEE Trans. Smart Grid, vol. 3,
no. 1, pp. 1225, Mar. 2012.
[16] A. Hawkes and M. Leach, Modelling high level system design and
unit commitment for a microgrid, Appl. Energy, vol. 86, no. 78,
pp. 12531265, Jul. 2009.
[17] I. Gradshteyn and I. Ryzhik, in Table of Integrals, Series, and Products,
A. Jeffrey and D. Zwillinger, Eds., 7th ed., New York, NY, USA: Elsevier
Academic Press, 2007.
[18] H. Feroze, Multi-agent systems in microgrids: Design and implementa-
tion, Master Thesis, Virginia Tech, 2009.
[19] L. Tao and C. Schwaegerl, Advanced architectures and control concepts
for more microgrids, EC Project, Tech. Rep. SES6-019864, 2009.
ZHAO et al.: SHORT-TERM OPERATION SCHEDULING IN RENEWABLE-POWERED MICROGRIDS
[20] M. Pipattanasomporn, H. Feroze, and S. Rahman, Securing critical loads
in a PV-based microgrid with a multi-agent system, Renew. Energy,
vol. 39, no. 1, pp. 166174, Mar. 2012.
[21] J. Carta, P. Ramrez, and S. Velzquez, A review of wind speed probability
distributions used in wind energy analysis, Renew. Sustain. Energy Rev.,
vol. 13, no. 5, pp. 933955, Jun. 2009.
[22] S. Roy, Market constrained optimal planning for wind energy conversion
systems over multiple installation sites, IEEE Trans. Energy Convers.,
vol. 17, no. 1, pp. 124129, Mar. 2002.
[23] F. Barbir and T. Gmez, Efciency and economics of proton exchange
membrane fuel cells, Int. J. Hydrogen Energy, vol. 13, no. 8,
pp. 10271037, 1997.
[24] F. Mohamed, Microgrid modelling and online management, Ph.D.
dissertation, Helsinki University of Technology, Helsinki, Finland, 2008.
[25] A. Azmy and I. Erlich, Online optimal management of PEM fuel cells using
neural networks, IEEE Trans. Power Del., vol. 20, no. 2, pp. 10511058,
Apr. 2005.
[26] J. L. Bernal-Agustn and R. Dufo-Lpez, Economical and environmental
analysis of grid connected photovoltaic systems in Spain, Renew. Energy,
vol. 31, no. 8, pp. 11071128, Jul. 2006.
[27] F. Mohamed and H. Koivo, Multiobjective optimization using modied
game theory for online management of microgrid, Eur. Trans. Electr.
Power, vol. 21, no. 1, pp. 839854, 2011.
[28] A. Saber and G. Venayagamoorthy, Plug-in vehicles and renewable energy
sources for cost and emission reductions, IEEE Trans. Ind. Electron.,
vol. 58, no. 4, pp. 12291238, Apr. 2011.
[29] R. Doherty and M. OMalley, A new approach to quantify reserve demand
in systems with signicant installed wind capacity, IEEE Trans. Power
Syst., vol. 22, no. 2, pp. 587595, May 2005.
[30] S. Tewari, C. J. Geyer, and N. Mohan, A statistical model for wind power
forecast error and its application to the estimation of penalties in liberalized
markets, IEEE Trans. Power Syst., vol. 26, no. 4, pp. 20312039,
Nov. 2011.
[31] S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge, U.K.:
Cambridge Univ. Press, 2004.
[32] D. P. Bertsekas, Nonlinear Programming, Belmont, MA, USA: Athena
Scientic, 1995.
[33] Wind energy database of white deer [Online]. Available: http://www.
windenergy.org/datasites/14-whitedeer/.
[34] W. Zhang, Branch-and-Bound search algorithms and their computational
complexity, University of Southern California Marina Delrey Information
Sciences Inst, Tech. Rep. ADA314598, May 1996.
[35] What is time complexity of genetic algorithm [Online]. Available: http://
wiki.answers.com/Q/What-is-time-complexity-of-genetic-algorithm.
[36] D. Lew and M. Milligan, The value of wind power forecasting, in Proc.
91st American Meteorological Society Annual Meeting, Seattle, WA, USA,
Jan. 2011.
[37] S. Arefar, Y. Mohamed, and T. El-Fouly, Supply-adequacy-based opti-
mal construction of microgrids in smart distribution systems, IEEE Trans.
Smart Grid, vol. 3, no. 3, pp. 14911502, Sep. 2012.
Binyan Zhao (S13) received the B.Sc. and M.Sc.
degrees in telecommunication from Beijing Universi-
ty of Posts and Telecommunications, Beijing, China,
in 2008 and 2011, respectively. She is currently
working toward the Ph.D. degree at the Department
of Electrical and Computer Engineering, University
of Victoria, Victoria, BC, Canada.
Her research interests include scheduling problems
in distribution power network and service restoration
issues in microgrid.
Yi Shi (S06M10) received the B.E. degree in tele-
communications from Beijing University of Posts
and Telecommunications, Beijing, China, in 2006. He
received the Ph.D. degree in electronic and computer
engineering from The Hong Kong University of
Science and Technology, Hong Kong, in August 2010.
From 2010 to 2012, he worked as a post-doctoral
fellow in the Department of Electrical and Computer
Engineering, University of Victoria. He is currently an
Engineer with Huawei Technologies, Beijing, China.
His work involves the 3GPP standarization of the
217
next-generation wireless mobile communication systems. His research interests
include cooperative communications, resource allocation and scheduling using
optimization and game theory, etc.
Xiaodai Dong (S97M00SM09) received the
B.Sc. degree in information and control engineering
from Xian Jiaotong University, Xian, China, the
M.Sc. degree in electrical engineering from National
University of Singapore, Singapore, and the Ph.D.
degree in electrical and computer engineering from
Queens University, Kingston, ON, Canada, in 1992,
1995, and 2000, respectively.
She is currently a Professor and Canada Research
Chair (Tier II) with the Department of Electrical
and Computer Engineering, University of Victoria,
Victoria, BC, Canada. Between 2002 and 2004, she was an Assistant Professor
with the Department of Electrical and Computer Engineering, University of
Alberta, Edmonton, AB, Canada. From 1999 to 2002, she was with Nortel
Networks, Ottawa, ON, where she worked on the base transceiver design of
third-generation mobile communication systems. Her research interests include
communication theory, radio propagation, cooperative communications, ultra-
wideband radio, and signal processing for communication applications.
Dr. Dong is an Editor for the IEEE TRANSACTIONS ON WIRELESS COMMUNICA-
TIONS, the IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, and the Journal of
Communications and Networks.
Wenpeng Luan (SM05) received the B.Sc. degree
from Tsinghua University, Beijing, China, in 1986,
the M.Sc. degree from Tianjin University, Tianjin,
China, in 1989, and the Ph.D. degree from Strathclyde
University, Glasgow, U.K., in 1999, all in electrical
engineering.
He was an Engineering Team Leader in the Smart
Metering and Infrastructure Program at BC Hydro,
Burnaby, BC, Canada. He is currently working at the
State Grid Smart Grid Research Institute, Beijing,
China. His special elds of interest include smart
metering, smart grid solutions, system planning, asset management, control
system cyber security, and utility advanced applications.
Jens Bornemann (M87SM90F02) received the
Dipl.-Ing. and the Dr.-Ing. degrees, both in electrical
engineering, from the University of Bremen,
Germany, in 1980 and 1984, respectively.
From 1984 to 1985, he was a Consulting Engineer.
In 1985, he joined the University of Bremen,
Germany, as an Assistant Professor. Since April
1988, he has been with the Department of Electrical
and Computer Engineering, University of Victoria,
Victoria, BC, Canada, where he became a Professor in
1992. From 1992 to 1995, he was a Fellow of the
British Columbia Advanced Systems Institute. In 1996, he was a Visiting
Scientist at Spar Aerospace Limited (now MDA Space), Ste-Anne-de-Bellevue,
Qubec, Canada, and a Visiting Professor at the Microwave Department, Univer-
sity of Ulm, Germany. In 2003, he was a Visiting Professor at the Laboratory for
Electromagnetic Fields and Microwave Electronics, ETH Zurich, Switzerland.
From 1997 to 2002, he was a codirector of the Center for Advanced Materials and
Related Technology (CAMTEC), University of Victoria. From 1999 to 2002, he
served as an Associate Editor of the IEEE TRANSACTIONS ON MICROWAVE THEORY
AND TECHNIQUES in the area of microwave modeling and CAD. From 2006 to 2008,
he was an Associate Editor of the International Journal of Electronics and
Communications. From 1999 to 2009, he served on the Technical Program
Committee of the IEEE MTT-S International Microwave Symposium. He
has coauthored Waveguide Components for Antenna Feed SystemsTheory and
Design (Artech House, 1993) and has authored/coauthored more than 250 tech-
nical papers. His research activities include RF/wireless/microwave/millimeter-
wave components and systems design, and eld-theory-based modeling of
integrated circuits, feed networks, ultrawideband technology, and antennas.
Dr. Bornemann is a Registered Professional Engineer in the Province of British
Columbia, Canada. He is a Fellow of the Canadian Academy of Engineering
(CAE) and serves on the editorial advisory board of the International Journal of
Numerical Modeling.