Ain Shams Engineering Journal (2015) xxx, xxx–xxx

Ain Shams University

Ain Shams Engineering Journal

www.elsevier.com/locate/asej
www.sciencedirect.com

A load shedding scheme for DG integrated

islanded power system utilizing backtracking
search algorithm
Aziah Khamis a,b,*, Hussain Shareef c, Azah Mohamed d, Zhao Yang Dong b

a
Faculty of Electrical Engineering, Universiti Teknikal Malaysia Melaka, Malaysia
b
School of Electrical and Information Engineering, University of Sydney, Australia
c
Electrical Engineering Department, College of Engineering, United Arab Emirates University, Al Ain, United Arab Emirates
d
Department of Electrical, Electronic and System Engineering, University Kebangsaan Malaysia, Malaysia

Received 24 September 2014; revised 30 September 2015; accepted 3 October 2015

KEYWORDS Abstract In a dispersed generation (DG) integrated distribution system, several technical issues
Optimal load shedding; should be resolved if the grid disconnects and forms an islanded system. The most critical challenge
Islanded system; in such a situation is to maintain the stability of the islanded system. The common practice is to
Voltage instability; reject several loads through a load shedding scheme. This study introduces a development of an
Voltage stability margin optimal load shedding scheme based on backtracking search algorithm (BSA). To handle this opti-
(VSM); mization problem, a constraint multiobjective function that considers the linear static voltage sta-
Backtracking search bility margin (VSM) and amount of load curtailment is formulated. It also handles the load priority
algorithm (BSA); and various operating conditions of DGs. The performance of the proposed load shedding scheme
Genetic algorithm (GA) was evaluated through an extensive test conducted on the IEEE 33-bus radial distribution system
with four DG units considering several scenarios such as load shedding under various operating
points and at various islands using the MATLABÒ software. Moreover, the effectiveness of the pro-
posed scheme was validated by comparing its results with those obtained using the genetic algo-
rithm (GA). The optimization results indicate that the proposed BSA technique is more effective
in determining the optimal amount of load to be shed in any islanded system compared with GA.
Ó 2015 Faculty of Engineering, Ain Shams University. Production and hosting 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/).

* Corresponding author at: Faculty of Electrical Engineering, Univer-

siti Teknikal Malaysia Melaka, Malaysia. Tel.: +60 65552256. 1. Introduction
E-mail addresses: [email protected] (A. Khamis), [email protected].
ae (H. Shareef), [email protected] (A. Mohamed), joe.dong@sydney.
edu.au (Z.Y. Dong). Considering the development of new technologies through the
Peer review under responsibility of Ain Shams University. years, the idea of immediately disconnecting all of the dis-
persed generations (DGs) to prevent equipment damage and
eliminate safety hazards is inapplicable. Moreover, some stan-
dards and regulations have been created to prevent islanding
Production and hosting by Elsevier hazards, and distribution network operator companies have
http://dx.doi.org/10.1016/j.asej.2015.10.001
2090-4479 Ó 2015 Faculty of Engineering, Ain Shams University. Production and hosting 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/).
Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
2 A. Khamis et al.

p k m q control, earthling or equivalent protection of the island opera-

Branch i tion, and resynchronization to the grid. Among these technical
hurdles, voltage and frequency control tends to occur more
frequently, of which load shedding is considered the most
Figure 1 Typical radial feeder of distribution system. effective technique to overcome the problem.
Generally, automatic load shedding has two types. The first
type is under-frequency load shedding (UFLS), which is
Start designed to rebalance load and generation within an electrical
island once the unbalanced system is created. The second type
Initialization is under-voltage load shedding (UVLS), which is utilized to
Pij~U(lowj,upj)
prevent local area voltage collapse and to directly respond to
the voltage condition in a local area. The UVLS scheme aims
Selection-I to shed load to restore reactive power relative to demand, to
oldPij~U(lowj,upj) prevent voltage collapse, and to contain a voltage problem
within a local area rather than allowing it to spread in a wide
Evaluate the population fitness, P
area. By contrast, automatic UFLS is designed for extreme
conditions to stabilize the balance between generation and
load after electrical island formation and to drop sufficient
ab~U(1,0) load to allow the frequency stabilization in the island. How-
ever, the UFLS is ineffective if instability or voltage collapse
No occurs within the island. Moreover, the most common factor
a<b
that contributes to power blackout is voltage instability [2].
Yes Thus, effective load shedding is crucial to prevent total system
oldP=P collapse. Improper load shedding would cause a high number
of blackouts.
oldP=permuting(oldP) Various load shedding schemes have been proposed by
researchers, in which the most applicable method is optimal
Mutation load shedding using computational intelligence techniques,
Mutant=P+F.(oldP-P) such as artificial neural network (ANN), adaptive neuro-
fuzzy inference system (ANFIS), fuzzy logic control (FLC),
genetic algorithm (GA), and particle swarm optimization
Generate mapping matrix, map
(PSO). For instance, [3] suggested controlling the voltage sta-
Crossover

No
bility during load shedding using FLC. The simulation results
Tij =Tij mapij=1? indicate that load shedding based on FLC can successfully sta-
Yes bilize and restore the system to the nominal value. However,
the principal limitation of this technique is that the rules of
Tij =Pij
FLC should be applied correctly depending upon the system
under study. The study in [4] demonstrated the application
of GA for determining the optimal load shedding scheme, with
Evaluate the fitness Ti
and without DGs at the network. This optimization aimed to
minimize the sum of curtailed load and system losses. Further-
Selection-II

more, the location and the amount of load to be shed in the

No
fitnessTi<Pi
power system can be determined based on the GA application
Pi =Pi
according to [5]. However, it is observed that the GA required
Yes a longer computation time in determining the amount of load
Pi =Ti shed, thus limiting its use for online application [6,7]. A new
algorithm for the steady-state load shedding strategy was pro-
posed in [8], in which an alliance algorithm considering the
More Yes
iteration ?
effect of demand priorities on the operation of the power sys-
tem during emergencies was introduced. Differential evolution
No is subsequently applied for anticipatory load shedding based
fitnessPbest = min (fitnessP) on voltage stability [9]. An application of bacterial foraging
algorithm optimization was likewise presented to evaluate
the optimal load shedding scheme with the objective of mini-
End
mizing the total power losses, voltage stability index value,
Figure 2 General flowchart of BSA. and total cost of load shed [10]. Although all of these tech-
niques can determine the optimal load shedding scheme,
extensive research is still required to enhance the performance
the primary duty to protect the network and its customers of computational intelligence techniques.
from the hazards [1]. The technical hurdles to achieve a safe Various voltage stability indicators, power losses, and
and smooth operation of islanded events are speed governor amounts of MW to be shed are used in optimization
response, range of operating power, voltage and frequency evaluation. Similar to other indicators, the static voltage

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
A load shedding scheme for power system 3

Start

Generate initial population

Pij ~U(lowj,upj )

Selection-I
oldPij ~U(lowj,upj )

Iteration = 1

BSA

Network Run load flow and calculate VSMsys

information VSMsys=min(VSM1,VSM2,VSM3,..,VSMk ) Iteration + 1

oldP = permuting
Constraints No Create and replace (oldP)
Fitness Evaluation

satisfied? current individual

Yes Selection-II

Calculate the fitness value

Next individual Evaluate fitness
f = max(VSMsys+Premaining load)

No Tij =Pij
Population = max ?

Crossover and Mutation

Yes

No
Iteration = max ?

Yes

Best Solution

End

Figure 3 Optimal load shedding scheme using BSA.

stability margin (VSM) proposed in [11] for voltage stability may cause non-optimal load shedding scheme. Accordingly,
assessment can be used in load shedding schemes. VSM can this study adopts an effective optimization algorithm known
be used to evaluate critical loads in an islanded system using as the backtracking search optimization algorithm (BSA).
the system voltage profile. Thus, the load shedding problem BSA is equipped with higher feasibility, solution quality, and
can be formulated as an optimization problem using VSM, convergence speed compared with the original GA and PSO
and an effective optimization technique can provide a reliable algorithms used in previous load shedding schemes. Therefore,
solution to the problem. Commonly used optimization tech- this study describes the development of the optimal load shed-
niques such as GA and PSO have limitations in terms of com- ding scheme based on BSA in an islanded distribution system.
putational time and pre-mature convergence. These limitations For this purpose, a multi-objective function is formulated

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
4 A. Khamis et al.

35
Table 1 Rated maximum power of DGs.
30
DG DG types Maximum active
power rating (MW) 25

Power (kW)
1 PV generator 0.03
20
2 Constant power generator 0.8
3 PV generator 0.6 15
4 Constant power generator 0.4
10

0
100 4:00 9:00 14:00 19:00
Hours
90
(a)
Load Level (%)

80 600

70 500

400

Power (kW)
60
300

50 200
4:00 9:00 14:00 19:00
Hours 100

Figure 4 Hourly load profile for individual loads. 0

4:00 9:00 14:00 19:00

Hours
considering the VSM and the amount of load to be shed. The
MATPOWER Newton–Raphson-based power flow algorithm (b)
in MATLABÒ is utilized to evaluate the formulated multiob-
Figure 5 Hourly PV power production: (a) DG1 and (b) DG3.
jective optimization problem. The proposed methods consider
various system constraints, such as load priority, voltage, and
power generation limits.
Table 2 Percentage load priority limits for the IEEE 33-bus
2. Tools and methods used in the proposed method radial distribution system.
Bus number Percentage (%) Bus number Percentage (%)
This section describes the overview of various tools and 1 0 18 34
methods (VSM index and BSA) that are used to develop the 2 34 19 60
proposed load shedding scheme for the islanded power system. 3 23 20 53
The first stage of the proposed method involves problem 4 64 21 20
formulation. The main concept behind the VSM is given in 5 15 22 50
Section 2.1. Meanwhile, the main steps of the BSA are summa- 6 43 23 4
rized in Section 2.2. 7 35 24 15
8 21 25 10
9 5 26 59
2.1. Voltage stability margin 10 21 27 2
11 0 28 28
VSM aims to evaluate the closeness of the system to voltage 12 52 29 15
13 11 30 55
collapse. Thus, VSM can be used as an indicator to obtain
14 47 31 25
the optimal load shedding scheme. It is derived from a typical
15 57 32 30
radial feeder of a distribution system as depicted in Fig. 1, in 16 61 33 3
which branch i is connected between buses k and m. Given 17 37
the magnitude and angle of these bus voltages, the loading
index (Li) of branch i can be expressed as follows [11–13]:

2
Vm
Li ¼ 2 cos dkm  1 ð1Þ The Li index shown in Eq. (1) estimates the maximum load
Vk
level of a single line section. Given its linear relationship, it can
where Vm represents the voltage at bus m, Vk represents the be utilized to represent the voltage stability at any loading level
voltage at bus k and dkm represents the angle between bus k of any line section in the system. Similar to other voltage sta-
and bus m. bility indices, Li also varies between unity (at no load) and zero

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
A load shedding scheme for power system 5

Load profile
Table 3 Overall power demand and supply in islanded Power generation
system. 4.0

Island Maximum Available Maximum 3.5

amount of load DG amount of DG
3.0
demand (MW) supply (MW)

Power (MW)
A 3.715 ALL DG 1.83 2.5
B 1.405 DG1, DG2 0.83 2.0
C 2.335 DG1, DG2, DG3 1.43
D 2.325 DG1, DG2, DG4 1.23 1.5

1.0

0.5
Table 4 GA and BSA parameter settings. 0.0
Parameter GA BSA 4:00 9:00 14:00 19:00
Hours
Population size 50 50
Maximum iteration 1000 1000 Figure 7 Daily load profile and power generation for island A.
Cross Probability 0.96 –
Mutation rate 0.08 –
Some loads in the system should be curtailed to reduce the
loading level of feeders and increase VSMsys to an acceptable
(at voltage collapse point). Meanwhile, the VSM of the feeder level. The optimum amount of the load to be shed from the
is considered as the product of loading indices of all of the fee- system can be determined by using a suitable optimization
der branches, which can be addressed as algorithm, such as BSA. The next section describes the main
Y BSA steps in the development of the proposed load shedding
VSM ¼ Li ð2Þ scheme.
i¼X
2.2. Backtracking search optimization algorithm
where X represents a set of branches constituting the feeder
(from source bus p to end bus q). Thus, the overall system
BSA is a new evolutionary algorithm (EA) proposed by [14] to
VSM (VSMsys) consisting of multiple feeders can be evaluated as
handle numerical optimization problems. The development of
VSMsys ¼ minðVSM1 ; VSM2 ; VSM3 ; . . . ; VSMs Þ ð3Þ BSA is based on random mutation strategy. The general
flowchart of BSA consists of five main processes, namely, ini-
where s is the number of feeders in the system. tialization, selection-I, mutation, crossover, and selection-II
DG3

DG3
DG4

DG4
32
33
26
27
28

29
30
31

29

30
26
27

28

31
32
33
23
24
25

23
24

25
DG1

DG1
EXTERNAL

EXTERNAL
GRID

GRID
1
2
3
4

5
6

7
8
DG2 9

13

15
10
11
12

14

16
17
18

1
2
3
4

5
6

7
8
DG2 9

13

15
10
11
12

14

16
17
18
Island B
21
20

22
19

21

22

Island A
19
20

(a) (b)
DG3
DG3

DG4

DG4
27

26
27
28
29
30
31

32
29

31

33
26

28

32
30

33

23

24

25
23
24
25

DG1
DG1

EXTERNAL
EXTERNAL

GRID
GRID

1
1

2
3
4

5
6

7
2
3
4

5
6

8
DG2 9
8
DG2 9

18
13

15
13

15

10
11
12

14

16
17
10
11
12

14

16
17

18

Island C Island D
21
20

22
21
22

19
19
20

(c) (d)

Figure 6 Single line diagram of islanded systems, (a) power island A, (b) power island B, (c) power island C, and (d) power island D.

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
6 A. Khamis et al.

Load demand before optimization

Table 5 Summary of load shedding performance at hour 9.00. Power generation
Load demand after optimization
Island Load Power Load Total 4.0
demand mismatch curtailment remaining
3.5
(MW) (%) (MW) load after
optimization 3.0
(MW)

Power (MW)
2.5
BSA GA BSA GA
2.0
B 0.974 16 0.171 0.222 0.803 0.752
1.5
C 1.619 37 0.655 0.7 0.964 0.919
D 1.612 25 0.454 0.529 1.158 1.083 1.0

0.5

0.0
4:00 9:00 14:00 19:00
(see Fig. 2). Similar to other heuristic optimizations, the first Hours
process of the BSA is the initialization of individual parame- (a)
ters to be optimized. This process can be expressed as
Load demand before optimization
Pij  Uðlowj ; upj Þ ð4Þ Load priority limit
Load demand after optimization
where Pij is the jth individual element in the problem dimen- 1.1
sion D that falls in ith position in a population dimension 1.0
N, U represents the uniform distribution, and up and low 0.9
respectively represent upper and lower boundaries. 0.8
The second process is selection-I. This stage aims to deter- Load profile (pu) 0.7
mine the search direction based on the historical population 0.6
oldP. The initial oldP is represented as 0.5
oldPij  Uðlowj ; upj Þ ð5Þ 0.4
0.3
However, oldP will be re-updated using Eq. (6) in each iter-
0.2
ation at the beginning through the if–then rule:
0.1
if a < b then oldP :¼ Pja; b  Uð0; 1Þ ð6Þ 0.0
4:00 9:00 14:00 19:00
where a and b represent random numbers between 0 and 1, and
Hours
:¼ represents the update operation. The update of oldP is then
(b)
completed by randomly changing the order of individuals in
oldP as shown in Eq. (7). The updated oldP acts as a memory Figure 8 Proposed load shedding scheme performance for island
in the BSA that helps guide the search direction. (a) generation and load mismatch and (b) optimum load profile
oldP :¼ permutingðoldPÞ ð7Þ with load priority limits.

After the oldP is updated, a trial population, T is subse-

quently generated through mutation and it is given by
The final stage of BSA is selection-II stage. At this point,
Mutant ¼ P þ F  ðoldP  PÞ ð8Þ the fitness of trial population T is evaluated and original pop-
where F is an algorithm-dependent parameter utilized to con- ulation P is updated using greedy selection.
trol the amplitude of the search direction. In this study, the
standard Brownian walk is used at the mutant stage, and it 3. Problem formulation
is given by F = 3. rand, where rand represents the random
value obtained from a standard normal distribution. Mean- A multiobjective function consisting of a static voltage stability
while, the final form of T is generated at the crossover stage margin is considered in searching for an optimal load shedding
that involves two major steps. The first step is to generate a in the islanded system. Thus, the operational constrains and
binary integer-value matrix (map) of size N  D using the fitness functions are presented in Sections 3.1 and 3.2, respec-
same if–then rule adopted for the update of oldP. At the sec- tively. Meanwhile, the application of the BSA for the optimal
ond stage, the individuals of T are manipulated using relevant load shedding scheme is summarized in Section 3.3.
individuals in P as shown in Eq. (9).
if mapi;j ¼ 1 then Ti;j :¼ Pi;j ð9Þ 3.1. Operation constraints

Then, the boundary condition of trial population, T is sub-

The optimal load shedding in the islanded system aims to
sequently checked and corrected using the following
enhance the voltage stability margin and voltage profile. How-
expression:
ever, the following constraints should be considered during
Ti;j ¼ rand:ðupj  lowj Þ þ lowj ð10Þ optimization:

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
A load shedding scheme for power system 7

2 Load demand before optimization

Load priority limit
Load demand after optimization
1.95
0.4
1.9
Fitness Value

1.85 0.3

Power (MW)
1.8 0.2

1.75
0.1
1.7

0.0
1.65 5 10 15 20 25 30
0 100 200 300 400 500 600 700 800 900 1000
Iteration Bus Number
(a) (a)

0.16
Load demand before optimization
BSA
Load priority limit 0.14 GA
Load demand after optimization
0.12
0.4
Power (MW) 0.1

0.3 0.08
Power (MW)

0.06
0.2 0.04

0.02
0.1
0 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

Bus Number
0.0
5 10 15 20 25 30 (b)
Bus Number
1.005
(b)
1.02
1
1.01

1
Voltage (pu)

0.995

0.99
Voltage (pu)

0.98 0.99

0.97
BSA
0.985 GA
0.96 Base case
BSA
0.95 0.98
0 5 10 15 20 25 30 35
0.94 Bus Number
0 5 10 15 20 25 30 35
(c)
Bus Number
(c) Figure 10 Performance of load shedding scheme at hour 9.00 for
island A (a) individual load demand after optimization using GA,
Figure 9 Performance of proposed load shedding scheme at (b) comparison of individual load demand after optimization
hour 9.00 for island A (a) convergence characteristic, (b) individ- between BSA and GA, and (c) comparison of voltage profile
ual load demand after optimization, and (c) voltage profile. obtained using BSA and GA.

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
8 A. Khamis et al.

Power flow balance: The total power generation during be maintained throughout the process in obtaining the opti-
optimization should be equal to the total consumption as mum load shedding scheme. These limits can be expressed
shown in Eq. (11). in terms of the inequality function as
X X X
Pgi  Pdi  Ploss ¼ 0 Spriority  Sli  Sl ð14Þ
X X X ð11Þ
Qgi  Qdi  Qloss ¼ 0 where Sli is the candidate value of the remaining load
power, Sl is the load at bus i before load shedding, and
where Pgi and Qgi are the generated active and reactive powers, Spriority is the load priority limit.
respectively, and Pdi and Qdi are the active and reactive powers Voltage stability margin limit: The VSMsys must be main-
consumed by the load, respectively. Ploss and Qloss are the tained at a certain limit to maintain the voltage profile
active and reactive power losses in the network, respectively. within the nominal value using Eq. (13). The limit of
Power flow limit: The apparent power Sl that is transmitted VSMsys can be given by
through branch l must not exceed the maximum thermal
0  VSMsys  1 ð15Þ
limit Slmax in a steady-state operation:
However, in practice, the voltage profile must meet the
Sl  Slmax ð12Þ standard values [15]. Thus, the limit of VSMsys can be
addressed as
Bus voltage stability: To prevent the voltage instability of
0:67  VSMsys  1 ð16Þ
the system, the bus voltage at each bus i must be maintained
around its normal value Vi, specified as [Vimin, Vimax], Power generator limit: The generator power, Pgen must be
where Vimin is the minimum permissible value of the volt- maintained at its maximum to provide all available power
age at bus i and Vimax is the maximum permissible voltage to support the system. The limit of Pgen can be given by
at bus i. These limits can be expressed in terms of the
inequality function as
Pgen ¼ Pmax ð17Þ
Vimin 6 Vi  Vimax ð13Þ
Practically, this deviation can reach up to 10% of the nom- 3.2. Fitness function
inal voltage value [13,15].
The fitness function aims to evaluate the optimal load shed-
Load shed limit: The permissible value of load that can be ding scheme in islanded systems on the basis of some indices.
curtailed in the system is limited by the load priority limit. The constraints of the problems during evaluation should be
The minimum amount of load that should be maintained fulfilled to obtain the best fitness function value. Thus, the
for each load is stored in a priority list. Thus, it should overall fitness function is formulated as

900

1.92 800

700
1.9

600
Iteration count

1.88
Fitness

500

1.86 400

300
1.84

200

1.82
100

GA BSA GA BSA
Algorithm Algorithm
(a) (b)
Figure 11 Performance comparisons of GA and BSA in obtaining optimal load shedding in island A at hour 9.00.

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
A load shedding scheme for power system 9

0.12
BSA
corresponds to the number of buses in the islanded system
GA under study. The Lfactor varies between unity and zero. This
0.1 Lfactor vector should be in the range of [Spriority, Sl]. The
optimization processes are repeated several times where the
0.08 maximum f is selected as the best fitness value. The load shed-
Power (MW)

ding scheme that corresponds to the maximum f is selected

0.06 where it generates the optimal amount of the remaining load
at that particular hour.
0.04 The VSMsys element in Eq. (18) maintains the load shed-
ding scheme, in which it should follow constraints in
0.02 Eqs. (13) and (16) to avoid the voltage collapse in the islanded
system. This element can evaluate the critical load in the
0 islanded system using the system voltage profile. Meanwhile,
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
the Premaining load element in Eq. (18) is utilized to ensure that
Bus Number
the amount of the remaining load is maximum, such that it
(a)
has the lowest amount of load to be shed in the islanded system
0.2 where it can fulfill constraints in Eq. (17).
0.18 BSA

0.16
GA 3.3. Application of BSA for optimal load shedding scheme
0.14
The BSA is used with MATPOWER power flow to determine
Power (MW)

0.12
the optimal load shedding scheme in the islanded distribution
0.1 system. Fig. 3 illustrates a schematic of the procedure involved
0.08 in solving the optimal load shedding scheme in the islanded
0.06 system based on BSA.
0.04
4. Test system description
0.02

0
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 22 23 24 In this study, the four DG units are modeled as constant power
Bus Number sources, in which various DG power injections depend on the
(b) type and hour of the day. The type of DG and the maximum
active power rating of each DG are depicted in Table 1. Figs. 4
0.12
and 5 show the individual load profiles and daily PV generator
BSA
GA
power consumption, respectively, in the load shedding study.
0.1
The 100% load level at hour 15.00 (Fig. 4) indicates the base
case bus power value obtained from the original IEEE 33-
0.08
bus radial distribution system with four DG units. Moreover,
Power (MW)

the load priority list that indicates the minimum power (in per-
0.06
centage) that must be maintained is presented in Table 2 for
the IEEE 33-bus radial distribution system with four DG
0.04
units. Any load cannot be curtailed from a bus with a 100%
limit. Meanwhile, all loads can be curtailed from a bus with
0.02
a 0% limit. The load priority limit for a given bus is not a fixed
value of power. It varies with the load demand of the bus at a
0
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 26 27 28 29 30 31 32 33 particular hour. In other words, the minimum amount of
Bus Number power (Spriority) that must be maintained at a given hour is
(c) the product of base case load power, percentage load level,
and priority limit. Spriority is used in the Lfactor vector as the
Figure 12 Comparison of individual load demand after opti- lower boundary, whereas hourly load demand (Sl) is used as
mization by BSA and GA at hour 9.00 for (a) power island B, (b) the upper boundary in optimization.
power island C, and (c) power island D. The simulations are based on possible island scenarios as
illustrated in Fig. 6. The figure shows that four possible
islanded systems can be formed for the IEEE 33-bus radial dis-
f ¼ maxðVSMsys þ Premaining load Þ ð18Þ
tribution system with four DG units. Moreover, the optimal
where f is the fitness function, VSMsys is the overall system volt- load shedding must be performed for each system. The overall
age stability margin and Premaining load is the total remaining load. maximum amount of load and available DG supply for each
In this study, BSA is utilized to obtain the optimal load islanded system is presented in Table 3.
shedding scheme in the islanded system. The solution set con- On the basis of the preceding system conditions, namely,
sidered in this optimization is the load shedding (Lfactor) vec- the load demand, load priority limits, and available power
tor. The Lfactor vector contains the amount of load permitted from generators at each hour, the proposed BSA-based opti-
to be shed for each bus where the dimension of Lfactor vector mum load shedding is performed for each islanded system as

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
10 A. Khamis et al.

1.01
Table 6 Performance of BSA and GA in terms of fitness,
1 VSM, and amount of load curtailment at hour 9.00.
Islanding Fitness VSM Load curtailment
0.99
(%)
Voltage (pu)

0.98 BSA GA BSA GA BSA GA

A 1.9301 1.8219 0.9301 0.9365 46 50
0.97
B 1.8877 1.8278 0.8858 0.8944 18 23
C 1.9213 1.8824 0.9213 0.9202 40 43
0.96
D 1.8978 1.7971 0.8978 0.9107 28 33
Base case
BSA
0.95 GA

0.94
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 5. Results and discussion
Bus Number
(a) 5.1. Optimal load shedding for island A using BSA
1.01
Island A is formed when the main circuit breaker at bus one in
1 the IEEE 33-bus radial distribution system with four DG units
is opened. Power island A contains 32 loads, totaling
0.99
3.715 MW real load power, which should be supplied by all
four DGs namely, DG1, DG2, DG3, and DG4, (Fig. 6(a)
Voltage (pu)

0.98
and Table 3). Fig. 7 shows the load demand and available
0.97 power generation for island A on an hourly basis. The figure
shows that the power mismatch between the load demand
0.96 and power generation is large, that is, 43–63% from valley
Base case
BSA
to peak daily load. To ensure that island A maintains its oper-
0.95 GA ation, the proposed optimal load shedding scheme is applied to
identify the location and amount of load to shed.
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 22 23 24 Fig. 8 illustrates the summary of the proposed load shed-
Bus Number ding scheme performance for power island A. Fig. 8(a) shows
(b) that the amount of remaining load after the optimization is
1.01
close to the amount of power generated during the hourly
operation. This result proves that the proposed optimal load
1 shedding scheme based on the BSA technique can decide the
optimal amount of load to be shed without cutting excessive
0.99 load from the system. The difference between power genera-
tion and load demand after optimization in Fig. 8(a) shows
Voltage (pu)

0.98
the power loss in the system (at hour 11:00–14:00). The power
0.97
loss is low when the PV generators DG1 and DG3 support the
system. Conversely, the power loss is high during evening and
0.96 early morning hours when PV DGs are inactive. Fig. 8(b)
Base case
shows that the load profile after the proposed load shedding
0.95 BSA
GA
scheme is higher than the load priority limits. The result indi-
cates that the proposed BSA-based load shedding scheme can
0.94
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 26 27 28 29 30 31 32 33 fulfill the requirement for load priority limit.
Bus Number The performance at hours 9.00 is analyzed to further eval-
(c) uate the effectiveness of the proposed load shedding scheme.
Fig. 8(b) shows that the load demand at hour 9.00 is
Figure 13 Comparison of voltage profile before and after load
0.69 pu, which is approximately 2.575 MW. However, the
shedding at hour 9.00 for (a) power island B, (b) power island C,
available power generation at this hour is only 1.414 MW,
and (c) power island D.
and the power mismatch is approximately 45%. This result
implies that 45% of the load must be curtailed from the system
to operate the system. After applying the optimal load shed-
ding scheme, 1.185 MW is curtailed, leaving only 1.390 MW
shown in Fig. 6. The performance of the proposed algorithm is as the total remaining load. The convergence characteristic
subsequently validated by comparing the performance of the for the proposed optimal load shedding scheme using the
load shedding scheme developed using the GA technique. BSA technique is shown in Fig. 9(a) for this hour. It shows
All of the optimization parameters are standardized to that BSA converges and finds the solution after 90 iterations.
compare the algorithms fairly. Table 4 shows the necessary Meanwhile, Fig. 9(b) shows that the optimal load shedding
parameter settings for both optimization techniques used in scheme can fulfill the entire load priority limit requirement.
this study. The scheme did not completely curtail the load from the buses

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
A load shedding scheme for power system 11

with low priority limits, such as buses 9, 11, 27, and 33. More- using GA with less amount of load curtailed in all of the
over, Fig. 9(c) shows that the voltage profile of most buses is islanded cases. Meanwhile, Fig. 12 shows that the GA-based
improved after optimization. For instance, buses 23 and 24 scheme curtails more load than the BSA-based scheme in the
(Fig. 9(c)) show that the amount of load to be curtailed is islanded systems in which the difference for islands B, C,
depending upon the priority load. However, the reduction of and D ranges from 1% to 18%, 1% to 24%, and 10%
the voltage profile at both of the buses is based on the objective to 31%, respectively. Moreover, an improvement in voltage
function and control variable as stated in Section 3. Thus, the profile at most buses can be observed after the optimization
result indicates that all of the bus voltages are now within the (Fig. 13). Slight differences exist in voltage profile because of
acceptable range (i.e., 0.98 pu to 1.01 pu) according to IEEE load shedding performed by BSA and GA. Some of the voltage
18-2002. magnitudes obtained using GA are higher than those obtained
using BSA. As previously mentioned, this difference in voltage
5.2. Optimal load shedding for island A using GA profile is attributed to the larger amount of load curtailment
by GA than by BSA in the system.
Using the similar procedures, the proposed optimal load shed- Table 6 shows the statistical results for fitness value, VSM,
ding scheme is simulated and tested with GA. This study aims and load curtailment at hour 9.00. For all of the island cases,
to compare and validate the performance of the BSA technique BSA has obtained the optimal load shedding results with a higher
in the proposed load shedding scheme. For this study, the per- amount of fitness value as indicated in bold. This table shows
formance at hours 9.00 is further analyzed. The performance that the VSM and the amount of load curtailment obtained using
of the load shedding scheme using GA at hour 9.00 is shown BSA are considerably lower than those obtained using GA. This
in Fig. 10. Fig. 10(a) reveals that the optimal load shedding result proves that the proposed optimal load shedding scheme
scheme using GA can fulfill the entire load priority limit require- based on BSA can decide the optimal amount of load to be shed
ment. After applying the optimal load shedding scheme, without cutting a substantial load from the system.
1.278 MW is curtailed, leaving only 1.297 MW as the total
remaining load. The total remaining load suggested by GA is 6. Conclusion
0.093 MW less than the amount calculated by the BSA tech-
nique. This result suggests that BSA is better than GA in obtain- This study describes a novel of optimal load shedding scheme
ing the optimal amount of remaining load, considering that the based on the BSA. The problem was to maximize the static
main objective of this study was to minimize the amount of load VSM and the total remaining load in the islanded system in
to be shed without cutting a substantial load from the system. order to stabilize the system and prevent voltage collapse. To
Fig. 10(b) reveals that GA curtails more load than BSA in the evaluate the performance of the proposed optimal load shed-
islanded system, in which the difference ranges from 0% to ding scheme technique, various evaluation techniques were
41%. Moreover, an improvement in voltage profile at most used. The performance evaluation method considered compar-
buses is observed after the optimization process by GA ative study between the conventional GA techniques. The opti-
(Fig. 10(c)). Slight differences in voltage profile emerge because mization results show that the proposed BSA technique is
of the load shedding performed by BSA and GA. Some voltage more effective in determining the optimal amount of load to
profiles obtained using GA are higher than those obtained using be shed in any islanded system compared with GA.
BSA. This improvement in voltage is due to a larger amount of
load curtailment by GA than by BSA in the system. Acknowledgments
To evaluate the effectiveness of the proposed algorithm, the
performance of the two optimization algorithms at hour 9.00 The authors wish to acknowledge the Ministry of Education,
for island A is evaluated (Fig. 11). Given their different search- Malaysia, and Universiti Teknikal Malaysia Melaka, Malaysia
ing patterns and convergence characteristics, the capabilities of for supporting this study and Universiti Kebangsaan Malaysia,
BSA and GA are examined with 15 repetitions of the optimiza- Malaysia for supporting this research financially under Grants
tion process. Moreover, the algorithms are compared in terms ETP-2013-044 and ERGS/1/2013/TK02/UKM/03/1.
of convergence rate (number of iterations required to con-
verge) and optimal solution quality (fitness value). Fig. 11(a) References
shows that BSA outperforms GA with higher fitness value.
[1] Ecconnect. Assessment of islanded operation of distribution
Fig. 11(b) shows that the BSA has better performance than
networks and measures for protection. Polym Contents 2001;18
GA in terms of convergence rates because the former is more
():830–77.
consistent than the latter with smaller inter-quartile range. [2] Laghari JA, Mokhlis H, Bakar AHA, Mohamad H. Application
of computational intelligence techniques for load shedding in
5.3. Optimal load shedding for other islanded systems power systems: a review. Energy Convers Manage 2013;75(August
):130–40.
Similar to the load shedding procedures adopted to island A, [3] Sallam AA, Khafaga AM. Fuzzy expert system using load
shedding for voltage instability control. In: Proceedings of the
optimization procedures are applied on the remaining island-
2002 large engineering systems conference on power engineering;
ing scenarios, namely, islands B, C, and D. The performance
2002. p. 125–32.
at hour 9.00 is further analyzed. The statistical results at hour [4] Malekpour AR, Seifi AR, Hesamzadeh MR, Hosseinzadeh N. An
9.00 for load demand, power mismatch, load curtailment and optimal load shedding approach for distribution networks with
total remaining load after the optimization using BSA and GA DGs considering capacity deficiency modelling of bulked power
are summarized in Table 5. The table shows that the proposed supply. In: Australasian universities power engineering conference
load shedding scheme using BSA performs better than that (AUPEC’08); 2008. p. 1–7.

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001
12 A. Khamis et al.

[5] Tarafdar Hagh M, Galvani S. A multi objective genetic algorithm Hussain Shareef received his B.Sc with hon-
for weighted load shedding. In: 2010 18th Iranian conference on ours from IIT, Bangladesh, MS degree from
electrical engineering; 2010. p. 867–73. METU, Turkey, and PhD degree from
[6] Al-Hasawi WM, Elnaggar KM. Optimum steady-state load- UTM, Malaysia, in 1999, 2002 and 2007,
shedding scheme using genetic based algorithm. In: 11th mediter- respectively. He is currently an associate pro-
ranean electrotechnical conference, 2002. MELECON 2002; 2002. fessor at the Electrical Engineering Depart-
p. 605–9. ment, United Arab Emirates University
[7] Rad BF, Abedi M. An optimal load-shedding scheme during (UAEU). His current research interests are
contingency situations using meta-heuristics algorithms with power system deregulation, power quality,
application of AHP method. In: 2008 11th international confer- artificial intelligence and power system distri-
ence on optimization of electrical and electronic equipment; 2008. bution automation. He is a member of IEEE.
p. 167–73.
[8] Calderaro V, Galdi V, Lattarulo V, Siano P. A new algorithm for
steady state load-shedding strategy. In: 2010 12th international Azah Mohamed received her B.Sc from
conference on optimization of electrical and electronic equipment; University of London in 1978 and M.Sc and
2010. p. 48–53. PhD from Universiti Malaya in 1988 and
[9] Arya LD, Singh P, Titare LS. Different evolution applied for 1995, respectively. She is a professor in the
anticipatory load shedding with voltage stability considerations. Department of Electrical, Electronic and
Electr Power and Energy Syst 2012;42:644–52. Systems Engineering, Universiti Kebangsaan
[10] Afandie WNEA, Rahman TKA, Zakaria Z. Bacterial foraging Malaysia. Her main research interests are
optimization algorithm for load shedding. In: IEEE 7th interna- power system security, power quality, artificial
tional power engineering and optimization conference intelligence and renewable energy systems.
(PEOCO2013). June 2013; p. 722–726. She is a senior member of IEEE.
[11] Haque MH. A linear static voltage stability margin for radial
distribution system. IEEE Power Eng Soc General Meet
2006;639798:1–6.
[12] Khamis A, Shareef H, Mohamed A, Bizkevelci E. An optimal Zhao Yang Dong (M’99-SM’06) obtained his
load shedding methodology for radial power distribution systems Ph.D. degree from the University of Sydney,
to improve static voltage stability margin using gravity search. J Australia in 1999. He is now Professor and
Teknol 2014;68(3):71–6. Head of School of Electrical and Information
[13] Khamis A, Shareef H, Mohamed A. Islanding detection and Engineering, the University of Sydney. His
load shedding scheme for radial distribution systems integrated immediate role was Ausgrid Chair and
with dispersed generations, IET Generation, Transmission and Director of the Centre for Intelligent
Distribution; 2015. Electricity Networks (CIEN), The University
[14] Civicioglu P. Backtracking search optimization algorithm for of Newcastle, Australia. He previously held
numerical optimization problems. Appl Math Comput Apr. academic and industrial positions with the
2013;219(15):8121–44. Hong Kong Polytechnic University and
[15] IEEE Std 18-2002. IEEE standard for shunt power capacitors. Transend Networks (now TASNetworks), Australia. His research
Institute of Electrical and Electronics Engineers; 2002. interest includes smart grid, power system planning, power system
security, renewable energy systems, electricity market, and computa-
tional intelligence and its application in power engineering. He is an
Aziah Khamis received her B.Eng. from editor of IEEE Transactions on Smart Grid, IEEE Transactions on
Universiti Putra Malaysia, Malaysia in 2006, Sustainable Energy, IEEE PES Letters and IET Renewable Power
M.Sc from Newcastle University, United King- Generation.
dom (UK) in year 2009 and PhD degree from
Universiti Kebangsaan Malaysia, Malaysia in
year 2014. She is senior lecturer in Universiti
Teknikal Malaysia Melaka. She is now pursu-
ing postdoctoral programme at The University
of Sydney, Australia. Her main research interests
include intelligent application of power system
study, distributed generation and microgrid.

Please cite this article in press as: Khamis A et al., A load shedding scheme for DG integrated islanded power system utilizing backtracking search algorithm, Ain
Shams Eng J (2015), http://dx.doi.org/10.1016/j.asej.2015.10.001