Electric Power Systems Research 128 (2015) 110

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Probabilistic coordination of microgrid energy resources operation

considering uncertainties
Walied Alharbi a , Kaamran Raahemifar b,
a
Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON, Canada N2L 3G1
b
Department of Electrical and Computer Engineering, Ryerson University, Toronto, ON, Canada M5B 2K3

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents probabilistic coordination of distributed energy resources (DERs) operation in an
Received 22 January 2015 islanded microgrid with consideration of the associated uncertainties. In doing so, a comprehensive
Received in revised form 27 May 2015 stochastic mathematical model is developed which incorporates a set of valid probabilistic scenarios
Accepted 8 June 2015
for the uncertainties of load and intermittency in wind and solar generation sources. The uncertainty
Available online 3 July 2015
is addressed through a combination of a stochastic optimization model and additional reserve require-
ments. The model also includes hourly interruption costs for a variety of customer types as a means of
Keywords:
determining the optimal probabilistic interruptible load whose reliability-based value is low enough to
Distributed energy resources
Microgrid
enable it to be shed if necessary. A case study is carried out using a benchmark microgrid; numerical
Smart grid results demonstrate that coordinated operation of DERs brings notable benets in terms of expected
Uncertainty operation costs and system security. This probabilistic coordination further reduces the consequences of
the expected power dispatch of controllable generators and hourly unserved power.
Crown Copyright 2015 Published by Elsevier B.V. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction interact in a smart microgrid in the presence of uncertainty. The

Driven by the need for realizing the smart grid concept and
the requirement for reliable integration of sustainable and green
energy in a decentralized fashion, microgrids have received signi-
cant attention from researchers over the last decade [1]. Microgrids
can be dened as medium or small power systems that are either
isolated and control clusters of local Distributed Generation (DG)
units and loads, or connected to the main grid as a means of supple-
menting supply requirements [1]. Energy resource scheduling in a
microgrid is different from that of a large power network due to its
size, power exchange with the main grid, and charging/discharging
of energy storage system (ESS) [2]. Indeed, controllable DGs in
a microgrid are much smaller than power systems, reecting an
easier switching operation and hence a more exible scheduling
[2]. However, todays microgrids are subject to operational chal-
lenges such as bidirectional power ows and instantaneous power
balance, mainly in the presence of excessive renewable energy
generation. Thus to provide reliable service, several controllable
and non-controllable distributed energy resources (DERs) must
Corresponding author. Tel.: +1 4169795000x6097; fax: +1 4169795280.
E-mail addresses: walharbi@uwaterloo.ca (W. Alharbi), kraahemi@ee.ryerson.ca
(K. Raahemifar).
primary goal of this paper is to enable such interaction under
uncertainty, with an emphasis on islanded microgrids.
Numerous studies and control topologies related to the micro-
grid energy management problem have been reported in the
literature [312]. Basu et al. [3] proposed a differential evolu-
tion method for reducing fuel costs and emissions in a microgrid
based on a heat-and-power combination. The proposed model
included constraints related to real-time balance, DER capacity lim-
its, and heat balance inequality. A single optimization objective
was employed based on consideration of the cost of emissions as a
penalty factor. Ahn et al. [4] described an Economic Dispatch (ED)
model that accounts for active power reserve in case of isolated
operations. In this model, the operation of interconnected micro-
grids is kept stable by sharing power among different sections or
areas, with the main limitation being the capacity of the feeders
connecting the areas. Ross et al. [5] presented a Knowledge-Based
Expert System (KBES) that includes an hourly discrete scheduling
algorithm for an isolated microgrid. The KBES optimizes schedules
by minimizing the use of dump loads through incorporating an ESS,
thereby reducing operation costs and emissions.
Conti et al. [6] presented an optimization procedure that enables
optimal dispatch of DGs and an ESS in a medium-voltage (MV)
islanded microgrid with the objective of minimizing both operation
costs and emissions. In this method, the proposed multi-objective

http://dx.doi.org/10.1016/j.epsr.2015.06.010
0378-7796/Crown Copyright 2015 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-
nd/4.0/).
2 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110

microgrid energy management under different modes of opera-

Nomenclature tion. The approach consists of two layers: the schedule layer which
provides an economical operating scheme; and the dispatch layer
Sets and indices
which builds on the rst layer and considers the other constraints
do demand variation downward
of the system, such as power ow and voltage limits. A smart
J, j set of indexes of the generating units
grid strategy has been proposed by Arabali et al. [8] for match-
K, k set of indexes of the time periods
ing renewable energy sources (RESs) with the controllable heating,
S, s set of indexes of the scenarios
ventilation, and air conditioning loads using a hybrid system that
up demand variation upward
combines RESs and ESS. A generic algorithm based optimization
approach and a two-point estimation method were used to mini-
Parameters
mize the cost and increase the efciency of the hybrid system. Using
BUP ,BDOWN maximum upward/downward variation (%)
a stochastic framework, the authors further studied [9] the optimal
Cmax ,Cmin max/min allowable energy stored in the ESS
sizing and reliability of the hybrid system. A pattern search opti-
Cst , CE initial/nal energy stored in the ESS
max mization method and a sequential Monte Carlo simulation were
PES maximum allowable charge/discharge limits
employed to minimize the system cost and stratify the reliability
Pjmax capacity of unit j (kW)
requirements. In a number of other models proposed for optimi-
Pjmin minimum power output of unit j (kW) zing the operation of microgrids [1014], DERs and loads are fully
D,R-C-I
Pk,s demand with DR for each type of customer in period controlled by the microgrid aggregator which is treated as a non-
k, considering scenario s (kW) prot agent. In these studies, the objective was either to minimize
PV
Pk,s power output of solar generation in period k and the operating costs of the entire microgrid or to maintain a balance
scenario s (kW) between the demand and local generation.
W
Pk,s power output of wind generation in period k and To the best of our knowledge, the variability and uncertainty
scenario s (kW) impacts of excessive renewable energy generation on the unit
SRk,s spinning reserve requirement in period k and sce- commitment decisions and real-time dispatch of a microgrid with
nario s (kW) controllable DGs in the presence of ESS, demand response (DR)
c, d charge/discharge efciency of the ESS and interruption loads have not been investigated before. Thus,
S probability of occurrence of scenario s this paper examines probabilistic coordination of DERs on micro-
grid operation considering the associated uncertainties and hourly
Variables interruptible loads for a variety of customers. The work presented
ETC expected total cost of the microgrid ($) in this paper can be summarized as follows:
UP,R-C-I
Pk,s demand increase in period k and scenario s, for
three types of customer (kW) (1) A comprehensive stochastic mathematical model has been
Do,R-C-I developed to enable operation interactions of DERs under
Pk,s demand decrease in period k and scenario s, for
three types of customer (kW) uncertainty in an islanded microgrid in order to mitigate the
Ck,s energy stored in ESS until period k, considering sce- variability and intermittency associated with large-scale inte-
nario s (kWh) gration of renewable energy generation.
CkDR,R-C-I hourly price responsive demand for three types of (2) The coordinated inuence of DERs on microgrids operation has
customer ($/kWh) been examined with respect to their independent presence and
CkInt,R-C-I time-varying interruption cost for three types of with consideration of the stochastic operational framework.
customer ($/kWh) (3) The hourly interruption cost for a variety of customers has been
Pj,k,s output power of committed unit j in period k, incorporated as a means to determine the optimal probabilis-
considering scenario s (kW) tic load interruption, if required. The effects of probabilistic
CH
Pk,s charging power of the ESS in period k, considering coordination of DERs on load interruptions have also been
investigated.
scenario s (kW)
Int
Pk,s power interruption in the load in period k and sce-
nario s (kW) The structure of the paper is as follows. Section 2 discusses the
DCH
Pk,s discharging power of the ESS in period k, consider- main approaches proposed in the literature for handling uncertain-
ing scenario s (kW) ties in energy management problem, followed by a description of
DEM
Pk,s variable demand with DR in period k and scenario s the uncertainty sets created for representing the load, wind, and
solar power forecasting errors addressed in this work. A compre-
(kW)
hensive mathematical model of the energy management problem
Vj,k binary variable: =1 if unit j is online in period k; = 0
for isolated microgrids is introduced in Section 3. The developed
otherwise
CH model was applied to a benchmark microgrid; the numerical results
Vk,s binary variable: =1 if the ESS is charging in period k
are presented and discussed in Section 4. The paper is concluded in
and scenario s; =0 otherwise
Section 5.
VkInt,R-C-I binary variable: =1 if power is interrupted in period
k and scenario s; =0 otherwise
DCH
Vk,s binary variable: =1 if the ESS is discharging in period 2. Managing uncertainty in the operational planning
k and scenario s; =0 otherwise problem

Two main types of uncertainty affect power system operations:

Optimal Power Flow (OPF) problem is solved using a niching evo-
lutionary algorithm capable of nding multiple optima and the
objective functions variations in their neighborhood. Jiang et al.
[7] proposed a novel double-layer coordinated approach that offers
outage of the generation units and departures from the forecasts.
The outage of generation units can lead to a supply shortage in a
system, which is usually met from both spinning and non-spinning
operating reserves. Departures from the forecasts resulting from
uncertain loads and the integration of RESs add additional
 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110 3

operational uncertainty. Wind and solar generation depend on of 24 h, as follows:

wind speed and solar irradiance, respectively, which are difcult to
predict accurately. Indeed, the effects of the uncertainty associated         CH 
with wind and solar generation are more signicant in microgrids ETC = s Cj Pj,k,s + SUj,k + CES Vk,s DCH
+ Vk,s
due to the high penetration levels of such resources. There are sS kK jJ kK
numerous reports of a variety of methods that take into account
the uncertainties arising from load, wind, and solar power fore-  
Int,R-C-I
casting errors [1517]. When it comes to managing uncertainty in + Pk,s CkInt,R-C-I + do,R-C-I
Pk,s CkDR,R-C-I (6)
the operational planning, we adopted the approaches employed by kK kK
Chen et al. [15] and Saber and Venayagamoorthy [16], i.e., inclu-
sion of reserve requirements and stochastic optimization models. where CkInt,R-C-I = fch (k) pc; CkDR,RCI = WDR fch (k) pc
As reported by Ruiz et al. [17], combining the aforementioned fch (k) represents the interruption cost hourly factor, pc desig-
two approaches leads to superior unit commitment (UC) policies nates the average interruption cost based on the sector customer
and enhanced representation of the uncertainties. Reserve require- damage function [18,19], and WDR is the weight factor of the effect
ments are determined from the forecasting errors set out by Chen of DR on customers.
et al. [15], which are represented in this paper as a percentage of The rst term in Eq. (6) indicates the expected operating cost
both the load demand and the wind and solar power output. With of the generators, which includes linearized fuel cost and the
stochastic optimization models, it is difcult to consider all con- start-up cost of unit j [20]. The readers may refer to [20] for an
tinuous states of the uncertainties. Therefore, for simplicity, in this explanation of the linearization form of the fuel cost and start-
paper a set of representative discrete states have been extracted. up cost of controllable generators. The second term denotes the
The discrete probability distribution sets of load (), wind (W ), expected operating cost of the ESS. The expected cost of power
and solar power (PV ) forecasting errors are represented as follows interruption is represented by the third term while the last term
[16]: of the equation signies the expected cost of the responsive
      demand.
D = PD1 , D
1
, PD2 , D
2
, . . . PDn , D
n
(1)
 1 1 2
 2
  q q

W = PW , W , PW , W , . . . PW , W (2) 3.1. Demand-supply balance
 1 1
 
2 2 m
 m
 
PV = PPV , PV , PPV , PV , . . . PPV , PV (3) This constraint ensures that the total generation meets the fore-
casted demand in period k, as presented in Eq. (7) which includes
In Eq. (1), PDi is the power demand associated with each state wind and photovoltaic (PV) generation, ESS, DR, and power inter-
i is the DEM,R-C-I
in the forecasting error probability distribution function, D ruption. The variable demand Pk,s represents the resulting
corresponding probability of uncertain load, and n is the number demand for each type of customer at hour k after demand shifting
of states in the discrete set. The discreet sets of wind and solar has taken place, and can be more or less than the original demand
D,R-C-I
power forecasting errors, Eqs. (2) and (3), are described similarly. Pk,s , as dened in (8) [21]:
The states probabilities should be equal to 1, as follows:

DEM,R-C-I Int,R-C-I

n 
q

m
W
Pj,k,s + Pk,s PV
+ Pk,s = Pk,s ESS
+ Pk,s + Pk,s s S (7)
i i i
D = W = PV =1 (4) jJ

i=1 i=1 i=1

where
The discrete sets in Eqs. (1)(3) are used to create a set of pos-
sible scenarios as given by (5), representing possible deviations ESS
Pk,s CH
= Pk,s DCH
Pk,s s S (8)
from the forecasted values of load, wind, and solar power. Each
scenario has a probability is which is equal to the product of the Int,R-C-I Int,R Int,C Int,I
Pk,s = Pk,s + Pk,s + Pk,s s S (9)
probabilities of the states corresponding to that scenario. For exam-
ple, if a scenario s, for a given state i, has xpv% more solar power DEM,R-C-I D,R-C-I Down,R-C-I
with probability PV , xw% less wind power with probability w , Pk,s = Pk,s UP.R-C-I
+ Pk,s Pk,s s S (10)
and xd% less load with probability d , then Ppvs (t) = (1 + xpv%)P (t),
pv
D,R-C-I D,R D,C D,I
PWs (t) = (1 xw%)P (t), P s (t) = (1 xd%)P , and their joint proba-
w d d Pk,s = Pk,s + Pk,s + Pk,s s S (11)
bility is pv W d .
UP.R-C-I
Pk,s UP.R
= Pk,s UP.C
+ Pk,s UP.I
+ Pk,s s S (12)

S

S

is = di w
i
pi v =1 S =nqm (5) Down.R-C-I
Pk,s Down.R
= Pk,s Down.C
+ Pk,s Down.I
+ Pk,s s S (13)
i=1 i=1

where S is the total number of scenarios. To prevent the customers activities from being shifted to the
The scenarios extracted and their corresponding probabilities next day, the demand variation must be balanced within the 24-h
are then included in the formulation of the microgrid energy man- operating horizon, as in (14). The maximum demand that can be
agement model. shifted from 1 h to another is given by (15) [21]:
 
UP,R-C-I Down,R-C-I
3. Comprehensive mathematical model of microgrid Pk,s = Pk,s s S (14)
energy management kK kK


Down,R-C-I

R-C-I
Bdp Pk,s
The mathematical model is formulated as a multi-scenario D,R-C-I
Mixed Integer Linear Programming (MILP) model. The models Pk,s s S (15)
R-C-I UP,R-C-I
BUP Pk,s
objective is to minimize the expected total cost (ETC) over a period
4 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110

1400
1200

Power, kW
1000
800
600
400
200
0
0 3 6 9 12 15 18 21 24
Time, h

Residential Industrial Commercial

Fig. 1. Load prole of the case study network.

3.2. Spinning reserve requirements of a microgrid and must be linearized to preserve the MILP formulations. Thus
Eqs. (24)(26) are added to represent the linearization form of the
This constraint ensures that the microgrid spinning reserve security action to ensure any excess load is interrupted in order to
requirements are met in period k. maintain power balance.
      
D,RCI
Pjmax Vj,k Pj,k,s + DCH Ck,s Cmin Rk,s + D Pk,s 1 if W
Pk,s + Pk,s
PV
+ Pjmax Vj,k D,R-C-I
SRks < Pk,s
Int,R-C-I
Vk,s = (23)
jJ jJ
0 Otherwise
W
+ W Pk,s PV
+ PV Pk,s s S (16)  
W
Pk,s PV
+ Pk,s + Pjmax Vj,k D,R-C-I
Pk,s Int,R-C-I
SRks (1 Vk,s ) s S (24)
where  D ,  W , and  PV are the forecasting error factors for load,
jJ
wind, and solar, respectively.
 
The rst term of Eq. (16) presents the net available capacity W
Pk,s PV
Pk,s Pjmax Vj,k D,R-C-I
+ Pk,s Int,R-C-I
+ SRks  Vk,s s S (25)
of all committed controllable DGs, and the second term denotes
jJ
the available capacity of the ESS. The last four terms of the equa-
Int,R-C-I int,R-C-I
tion represent the spinning reserve requirements. In fact, the last Pk,s  Vk,s s S (26)
three terms are the extra reserves added to mitigate the effects of
uncertainties on the load, wind, and solar power forecasts. 4. Model scenarios, results, and analysis
Each controllable generator is also subject to its own operating
constraints, such as ramp-up/ramp-down, minimum up and down 4.1. System under study
times, generation output limits, and logical and initial constraints
[20]. The model was validated using a benchmark microgrid consist-
ing of 12 controllable DGs, a wind turbine (WT), a PV system, and an
3.3. Energy storage systems ESS. The total installed capacity in the microgrid system is 4 MW,
with a renewable penetration level of 49%. The installed capacities
The following equations represent the operational constraints of controllable DGs, WT, and PV systems are 2.04 MW, 0.52 MW
of the ESS [15]: (4 0.14 MWp), and 1.44 MW (4 0.14 0.36 MWp), respectively
Power discharge and charge limits: [22]. The 24-h load prole for the microgrid, which includes the
DCH
Pk,s max
PES DCH
Vk,s sS (17) commercial, residential, and industrial loads derived from [23], is
shown in Fig. 1. Fig. 2 shows the hourly PV output power prole.
CH
Pk,s max
PES CH
Vk,s s S (18) The power output formulation of wind generation can be expressed
as follows [22,24]:
Discharging and charging dynamics:
DCH
Ck+1,s = Ck,s dk Pk,s CH
/DCH + dk Pk,s CH s S
0 0 Vm V1
(19)

aV 4 + b V 3 + cV 2 + dVm + e V1 Vm Vr
m m m
ESS operational end points and energy storage limits: out
Pw = (27)

Prated Vr Vm Vcut-out
C0,s = CSt , CK,s CE s S (20)


0 Vm Vcut-out
Cmin Ck,s Cmax s S (21)
where Vm , V1 , Vr , and Vcut-out are the wind speed, cut-in speed,
Coordination of charging and discharging operation:
speed at the rated power, and cut-out speed, respectively. Table 1
DCH
Vk,s DCH
+ Vk,s 1 s S (22)

3.4. Load interruption

In order to protect the systems security, load interruption is

invoked in cases of excess demand. Moreover, additional load inter-
ruption should take place to free up a portion of the capacity of the
controllable unit to provide the required spinning reserve, mainly
in cases where ESSs are not included in the system. Eq. (23) presents
load interruption decision variable which is only utilized when
security risks are present. However, this condition is nonlinear Fig. 2. PV output prole for the microgrid system.
 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110 5

16

14

12

Wind speed, m/s

10

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time, h

Fig. 3. Wind speed prole for the wind power output.

Table 1 developed comprehensive mathematical model was solved using

Wind plant data.
the CEPLX solver in the GAMS platform [27].
Prated V1 Vr Vcut-out a b c d e

140 3.00 15.01 17.00 0.015 0.33 0.9 2.10 7.10 4.2. Results and analysis

The objective of the microgrid operator is to minimize the total

presents the wind plant data, and Fig. 3 shows the wind speed pro-
le. The microgrid operator is assumed to own and operate the
ESS. The average ESS operating cost is estimated to be 0.54$/kWh,
based on two operating cycles per day, with 1200$/kWh of invest-
ment at an 8% annual rate of return for a life span of 3000 cycles
[25]. The ESS considered to be associated with this microgrid is a
single-layer Vanadium Redox Flow Batteries [5]. For the ESS, PES max
is 150 kW, Cmax is 900 kWh, CS and CE are 0.5 p.u., and the charging
and discharging efciencies are 0.85 [5].
One-third of the load is assumed to be critical. The spinning
reserve requirement for each hour is considered to be 10% of the
critical load for that hour. The shiftable demand is assumed to be
15% for each type of customer. The hourly average interruption
costs for residential, commercial, and industrial customers are
0.482$/kW, 8.552$/kW, and 9.085$/kW, respectively [18]. The
typical hourly factors for the interruption costs associated with
the three types of customers are shown in Fig. 4. All customers
are assumed to be informed about demand shifting in advance,
and since demand shifting has less effect than shedding, the
weight factor for a demand response is assumed to be 0.5. The
constant  used in the linearized Eqs. (18)(20) is a large arbitrary
number to ensure those equations are satised. The value of 
is 10,000. A normal ve-state discrete probability distribution of
the forecasting errors related to load, wind, and solar power is
used to represent the uncertainty, as given in Table 2 [16]. A total
of 125 scenarios were thus generated using a two-stage scenario
tree. The forecasting error factors applicable to the spinning
reserve constraint were adopted from the paper by Rahimiyan
et al. [14]:  D = 0.03,  W = 0.13, and  PV = 0.09. Data of pollution
contaminants of different generators can be found in [26]. The
Table 2
The discreet probability distribution of forecasting errors.
Load Wind
(%) Prob. () (%) Prob. ()
2.5 0.05 2.5 0.10
1 0.15 1 0.15
0 0.60 0 0.50
+2 0.15 +1 0.15
+3 0.05 +2.5 0.10
expected operation cost by optimizing the DERs schedules which
include controllable and uncontrollable DGs, ESS, DR and inter-
ruption loads. In order to demonstrate the effectiveness of the
proposed framework, the following four cases were examined and
run in this study:
Case-1 (base case): RESs, controllable DGs, and interruptible
loads.
Case-2 (ESS case): RESs, controllable DGs, interruptible loads, and
ESS.
Case-3 (DR case): RESs, controllable DGs, interruptible loads, and
DR.
Case-4 (coordination case): RESs, controllable DGs, interruptible
loads, ESS, and DR.
The rst case ignores the ESS and DR and focuses on the impact
of the variability and uncertainty arising from renewable energy
generation, and the amount of load interruptions needed to ensure
system security and reliability. The ESS is included in Case-2 while
DR is considered in Case-3. A combination of both ESS and DR is
represented by Case-4. Table 3 represents the expected simulation
results for each case for an isolated microgrid. In Case-1, when the
system load exceeds the total generation capacity, the expected
excess load is interrupted, with the timing and amount of the inter-
ruption being chosen according to its hourly interruption value. An
additional load interruption also frees up a portion of the control-
lable DG to provide the required spinning reserve in the system,
primarily in Case-1 and Case-3. This load interruption is reduced
as additional exibility, in terms of smart grid components, and is
introduced into the microgrid. Thus, the fourth case entails the least
amount of load interruption among all cases. In fact, Case-4 is con-
sidered as an economic and reliable case (lower total operating cost
and fewer load interruptions) in which effectiveness and efciency
are gained by coordinating the operation of all smart grid energy
Solar resources of the microgrid, with consideration of the uncertainties
(%) Prob. () involved.
2 0.05
Table 4 shows changes in the expected unit commitment (UC)
1.5 0.15 decisions from Case-1 to Case-4, over a period of 24 h. The gray
0 0.60 cells are those changing from ON-state to OFF-state and vice versa.
+1.5 0.15 For instance, the decision of DG2 changed from off-state (Case-1)
+2 0.05
to on-state (Case-4) during hours 610, while the decision of DG5
6 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110

2.5

Interruption cost-hour factor

2

1.5

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time, h

Residential Commercial Industrial

Fig. 4. Interruption cost hourly factors for the three types of customers during 1 day.

Table 3
Cost components and emissions for islanded microgrids.

Case Expected operating Expected Expected DR ETC ($) Expected emissions

cost ($) interruption cost cost ($) (Ib/kWh)
($)

1 4164 2290 0 6454 4101

2 4323 584 0 4907 4223
3 4200 772 191 5163 4161
4 4379 292 120 4791 4284

Table 4
Unit commitment decisions for an isolated microgrid.

Hour 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Generator

DG1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
DG2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
DG3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1
DG4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1
DG5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 0
DG6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 0
DG7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 0
DG8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 0
DG9 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 0
DG10 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 0
DG11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 0
DG12 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 0

changed from on-state (Case-1) to off-state (Case-4) during hours of controllable DGs during the worst scenario, in which the load
116. It can be noticed that probabilistic coordination of DERs oper- prole exhibits the highest positive values compared to the pre-
ations makes UC decisions of controllable DGs smooth and without fect forecasted values, while the wind and solar energy generation
any uctuation, and thus reduces the total expected operation cost proles have the highest negative values relative to their prefect
by reducing the starting up and shutting down costs of controllable forecasted values. It can be observed that coordinating the DERs,
DGs. as in Case-4, further attens the net demand and hence reduces
Fig. 5 illustrates the effects of the variability and uncertainty the impact of variability and uncertainty on the real-time dispatch
associated with wind and solar generation on the power dispatch of controllable DGs. During peak times, in Case-2 and Case-4, the
DGs are dispatched to their fully installed capacities since the ESS
is considered to be providing the required spinning reserve in the
100 system.
Expected power dispatch of controllable

DGs Alone DGs with ESS

DGs, % of their installed capacities

90
DGs with DR DER Furthermore, the power dispatch of controllable DGs for 2 of the
80
125 scenarios, i.e., the best and worst scenarios, are compared with
70
that of the prefect scenario. The worst scenario has been dened
60
above. In the best scenario however, the load prole has the high-
50
est negative values compared to the prefect forecasted values and
40
30
the wind and solar energy generation proles exhibit the high-
20
est positive values relative to their perfect forecasted values. The
10 power dispatch of controllable DGs for the worst and best scenar-
0 ios and the perfect forecast for Case-1 and Case-4 are presented in
0 3 6 9 12 15 18 21 24
Figs. 6 and 7, respectively. Fig. 6 shows the impact of high variabil-
Time, h
ity and uncertainty of renewable power generation on the power
Fig. 5. Optimal dispatch of controllable DGs (worst scenario). dispatch of controllable DGs, mainly during the best and worst
 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110 7

Expected Power dispatch deviations of

100
75
50

controllable DGs, kW
25
0
-25 0 3 6 9 12 15 18 21 24

-50
-75
-100
Time, h

Highest Negative Forecast from Perfect Forecast (Worst Scenario)

Highest Positive Forecast from Perfect Forecast (Best Scenario)
Perfect Forecast

Fig. 6. Dispatch deviations from the perfect forecast for Case-1.

Expected Power dispatch deviations of

180
150
120
Controllable DGs, kW

90
60
30
0
-30 0 3 6 9 12 15 18 21 24
-60
-90
-120
-150
-180
Time, h

Highest Positive Forecast from Perfect Forecast (Best Scenario)

Highest Negative Forecast from Perfect Forecast (Worst Scenario)
Perfect Forecast

Fig. 7. Dispatch deviations from the perfect forecast for Case-4.

scenarios. In the best scenario, the power dispatch of controllable
DGs reduces with respect to the prefect forecast. In the worst sce-
nario however, the power dispatch of controllable DGs increases.
In contrast probabilistic coordination of DERs either shrinks the
gap between the worst/best scenario and the prefect forecast in
order to reduce the renewable variability and uncertainty effects
on power dispatch of controllable DGs, or widens the gap during
off-peak periods to charge the ESS and/or induce optimal DR to
use during on-peak hours (Fig. 7), and therefore further reduces
the total expected operation costs. It is worth noting that the gap
between the best/worst scenario and prefect scenario is created
from forecasting errors of load, wind and solar energy generation
proles.
Fig. 8 shows ESS charging/discharging behaviors and state of
charge of the ESS during the worst scenario, both alone and in coor-
dination with the DR. Similarly, DR behaviors with and without ESS
are illustrated in Fig. 9. Fig. 8a shows the charging and discharging
cycles of ESS which follows the load prole when DR is not consid-
ered. This means charging ESS during off-peak and discharging it
during on-peak periods. However, this behavior is changed when
the ESS is in coordination with DR, which in this case follows the
original load plus the deferrable loads. Fig. 8b represents the state
of charge of the ESS according to its charging/discharging behaviors
over a period of 24 h. Fig. 9 shows that the ESS has no signicant
impact on DR since the DR is derived primarily from its respon-
sive price, and hence occurs when the price is low. However, the
opposite is not true because the DR can play a signicant role in the
amount of discrepancy between the expected power generation
and load and thus it will have a signicant effect on ESS charging
and discharging behaviors.
Since the hourly interruption cost is not as high for residen-
tial customers as for commercial or industrial customers, when an
excess demand occurs the residential loads are interrupted rst to
ensure system security. The expected hourly interruptible loads for
residential customers during the worst-case scenario are shown
in Fig. 10. The optimal probabilistic load interruption is clearly
induced, mainly during peak times, and varies noticeably from case
to case, depending on the condition and exibility of the system. It
is worth noting that, to ensure system security in Case-1, not only
the residential load, but also the commercial load is interrupted
(Fig. 11).
A sensitivity analysis is carried out to assess the impact of higher
forecasting errors of renewable energy generation and load on
microgrid operations and system security. The higher uncertainties
are obtained by assigning a large standard deviation to the probabil-
ity density functions of the forecasting errors. The new distribution

Table 5
Cost components for islanded microgrids considering higher forecasting errors.

Case Expected operating Expected Expected DR ETC ($) Expected emissions

cost ($) interruption cost cost ($) (Ib/kWh)
($)

1 4181 2407 0 6588 4123

4 4459 265 80 4804 4287
8 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110

Charging/Dsichrging of the
75

50

25

ESS,,%
0
0 3 6 9 12 15 18 21 24
-25

-50 ESS alone

-75 ESS in Coordination with DR
-100 Time, h

b
100
SOC of the ESS, %

75

50

SOC of the ESS alone

25
SOC of the ESS in Coordination with DR
0
0 3 6 9 12 15 18 21 24
Time, h

Fig. 8. (a) ESS charging/discharging behavior alone and in coordination with the DR during the worst scenario. (b) State of charge of the ESS alone and in coordination with
the DR during the worst scenario.

80
DR alone DR in Coordination with ESS
60
Expected responsive loads, kW

40

20

0
0 3 6 9 12 15 18 21 24
-20

-40
Time, h
-60

-80

Fig. 9. DR behavior with and without the ESS during the worst-case scenario.

functions are generated using Table 2, by multiplying the forecast- expected operation cost of the coordination case (Case-4) is still less
ing errors by 1.75 and maintaining the same probability. Table 5 than those of Case-2 and Case-3 (Table 3) where higher forecasting
presents the expected simulation results for the base case and coor- errors were not considered. This shows probabilistic coordination
dination case of the isolated microgrid. It can be observed that the of DERs operation reduces the risk in scheduling and increases the

450
Expected Interruptible Loads, kW

400
350
300
250
200
150
100
50
0
0 3 6 9 12 15 18 21 24
Time, h
Case-1 Case-2 Case-3 Case-4

Fig. 10. Optimal interruptible residential loads for each case during the worst-case scenario.

Expected Interruptible Loads, kW
Table 6
Model statistic for each case.
MILP optimization model
Case
Generation time (s)
Execution time (s)
Elapsed time (min)
Iteration count for convergence
exibility of microgrid in mitigating the effects of higher forecasting
errors with minimum cost.
Table 6 shows the generation time, execution time, elapsed
time, and iteration count for convergence for each of the four cases
of the developed model. The execution time increases when the
ESS is included in the model, and thus a decomposition approach
is required to allow the proposed model to be solved within the
desired window of time to make it suitable for real-time applica-
tions.
It is worth noting that this paper focuses on the systems
steady-state behavior and ignores its dynamic behavior. However,
the microgrid operator should be capable of handling dynamic
issues since the system has enough small generation units with
high operational exibility in terms of ramping-rates, minimum-
up/minimum-down times, and start-up/shut-down times to
respond quickly in order to mitigate any transient issue. Fur-
thermore, the proposed framework enhances system exibility by
coordinating operations of DERs and thus makes the system further
capable of handling dynamic matters with the least operation cost,
as illustrated in the steady-state study.
5. Conclusions
The signicant increase in penetration of renewable energy
generation is expected to affect the operational aspects of power
systems and more specically isolated microgrids that mainly rely
on renewable energy sources. Indeed, the issue of UC schedules
and real-time dispatch of a microgrid with controllable DGs is more
complicated in the presence of high wind and solar generation pen-
etration. To address these operational challenges, it is necessary
to manage the variability and uncertainty associated with these
energy sources. It has become clearer than ever that a exible
microgrid is vital. In this paper, a comprehensive stochastic mathe-
matical model has been developed to enable operation interaction
of DERs under uncertainty in islanded microgrids. Hence prob-
abilistic coordination of DERs on microgrid operations has been
examined with respect to their independent presence and with
consideration of the hourly interruption cost for residential, com-
mercial, and industrial customers in order to determine the optimal
W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110 9
150
120
90
60
30
0
0 3 6 9 12 15 18 21 24
Time, h
Fig. 11. Optimal interruptible commercial loads for Case-1 during the worst-case scenario.
probabilistic interruptible load, if required. The simulation results
show that coordinating all smart grid energy resources signicantly
enhances energy management of the microgrid and further reduces
the expected total operation cost and helps maintain grid balance
1 2 3 4
3.93 4.07 3.75 4.16 under high penetration of RESs, especially when higher forecasting
7.47 8.22 7.55 8.51 errors are considered. Furthermore, although DR behavior may not
1.06 14.33 0.73 13.61 be signicantly affected when coordinated with an ESS, the reverse
78,544 262,740 76,672 283,383
is not true. DR may impact not only the ESS charging/discharging
behavior, but also its optimal sizing. The aforementioned raises
an interesting research question worthy of future investigation,
especially with respect to the possible implementation of islanded
microgrid systems in smart distribution networks.
Acknowledgements
This research was supported by funding from the Saudi Gov-
ernment to Walied Alharbi and from The Natural Sciences and
Engineering Research Council of Canada (NSERC) to Kaamran Raa-
hemifar.
References
[1] P. Varaiya, F. Wu, J. Bialek, Smart operation of smart grid: risk-limiting dispatch,
Proc. IEEE 99 (January (1)) (2011) 4057.
[2] C.A. Hernandez-Aramburo, T.C. Green, N. Mugniot, Fuel consumption mini-
mization of a microgrid, IEEE Trans. Ind. Appl. 41 (2005) 673681.
[3] A.K. Basu, A. Bhattacharya, S. Chowdhury, S. Chowdhury, Planned scheduling
for economic power sharing in a CHP-based micro-grid, IEEE Trans. Power Syst.
27 (2012) 3038.
[4] S. Ahn, S. Nam, J. Choi, S. Moon, Power scheduling of distributed generators for
economic and stable operation of a microgrid, IEEE Trans. Smart Grid 4 (2013)
398405.
[5] M. Ross, R. Hidalgo, C. Abbey, G. Joos, Energy storage system scheduling for an
isolated microgrid, IET Renew. Power Gener. 5 (2011) 117123.
[6] S. Conti, R. Nicolosi, S. Rizzo, H. Zeineldin, Optimal dispatching of distributed
generators and storage systems for MV islanded microgrids, IEEE Trans. Power
Deliv. 27 (2012) 12431251.
[7] Q. Jiang, M. Xue, G. Geng, Energy management of microgrid in grid-connected
and stand-alone modes, IEEE Trans. Power Syst. 28 (3) (2013) 33803389.
[8] Arabali, M. Ghofrani, M. Etezadi-Amoli, M.S. Fadali, Y. Baghzouz, Genetic-
algorithm-based optimization approach for energy management, IEEE Trans.
Power Deliv. 28 (January (1)) (2013) 162170.
[9] Arabali, M. Ghofrani, M. Etezadi-Amoli, M.S. Fadali, Stochastic performance
assessment and sizing for a hybrid power system of solar/wind/energy storage,
IEEE Trans. Sustain. Energy 5 (April (2)) (2014) 363371.
[10] D. Nguyen, L. Le, Optimal energy trading for building microgrid with electric
vehicles and renewable energy resources, in: Proc. IEEE PES Innov. Smart Grid
Technol. (ISGT), Washington, DC, USA, February, 2014, pp. 15.
[11] R.P. Behnke, et al., A microgrid energy management system based on the rolling
horizon strategy, IEEE Trans. Smart Grid 4 (June (2)) (2013) 9961006.
[12] D. Nguyen, L. Le, Optimal bidding strategy for microgrids considering renew-
able energy and building thermal dynamics, IEEE Trans. Smart Grid 5 (July (4))
(2014) 16081620.
[13] B. Zhao, Y. Shi, X. Dong, W. Luan, J. Bornemann, Short-term operation sched-
uling in renewable-powered microgrids: a duality-based approach, IEEE Trans.
Sustain. Energy 5 (January (1)) (2014) 209217.
10 W. Alharbi, K. Raahemifar / Electric Power Systems Research 128 (2015) 110
[14] M. Rahimiyan, L. Baringo, A.J. Conejo, Energy management of a cluster of inter-
connected price-responsive demands, IEEE Trans. Power Syst. 29 (March (2))
(2014) 645655.
[15] S. Chen, H.B. Gooi, M. Wang, Sizing of energy storage for microgrids, IEEE Trans.
Smart Grid 3 (2012) 142151.
[16] A.Y. Saber, G.K. Venayagamoorthy, Resource scheduling under uncertainty in a
smart grid with renewables and plug-in vehicles, IEEE Syst. J. 6 (2012) 103109.
[17] P.A. Ruiz, C.R. Philbrick, E. Zak, K.W. Cheung, P.W. Sauer, Uncertainty man-
agement in the unit commitment problem, IEEE Trans. Power Syst. 24 (2009)
642651.
[18] I. Bae, J. Kim, C. Singh, Optimal operating strategy for distributed generation
considering hourly reliability worth, IEEE Trans. Power Syst. 19 (1) (2004)
287292.
[19] S. Junlakarn, M. Ilic, Distribution system reliability options and utility liability,
IEEE Trans. Smart Grid 5 (September (5)) (2014) 22272233.
[20] M. Carrin, J.M. Arroyo, A computationally efcient mixed-integer linear for-
mulation for the thermal unit commitment problem, IEEE Trans. Power Syst.
21 (2006) 13711378.
[21] K. Dietrich, J.M. Latorre, L. Olmos, A. Ramos, Demand response in an isolated
system with high wind integration, IEEE Trans. Power Syst. 27 (2012) 2029.
[22] T. Logenthiran, D. Srinivasan, A.M. Khambadkone, Multi-agent system for
energy resource scheduling of integrated microgrids in a distributed system,
Electr. Power Syst. Res. 81 (2011) 138148.
[23] A. Tsikalakis, N. Hatziargyriou, Centralized control for optimizing microgrids
operation, IEEE Trans. Energy Convers. 23 (2008) 241248.
[24] T. Logenthiran, D. Srinivasan, Short term generation scheduling of a microgrid,
in: IEEE TENCON, 2009, pp. 16.
[25] K. Divya, J. stergaard, Battery energy storage technology for power
systemsan overview, Electr. Power Syst. Res. 79 (2009) 511520.
[26] C. Volume II, Uncontrolled Emission Factor Listing For Criteria Air Pollutants,
July 2001. Prepared by: Eastern Research Group, Point Sources Committee and
Emission Inventory Improvement Program, located at http://www.epa.gov/
ttnchie1/eiip/techreport/volume02/index.html
[27] GAMS Development Corporation, General algebraic modeling system (GAMS)
Software, 2012 http://www.gams.com