Electrical Power and Energy Systems 63 (2014) 862–877

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Anticipatory load shedding for line overload alleviation using Teaching

learning based optimization (TLBO)
L.D. Arya a, Atul Koshti b,⇑
a
Electrical Engg. Department, Shri G.S. Institute of Technology and Science, Indore, M.P., India
b
Electrical Engg. Deptt., Swami Vivekanand College of Engineering, Indore, M.P., India

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

Article history: This paper presents a load shedding algorithm for alleviating line overloads employing Teaching learning
Received 1 July 2013 based optimization (TLBO). The buses are selected for load shed based on the sensitivity of severity index
Received in revised form 9 June 2014 with respect to load shed. Load shed is based on the next interval predicted load which could cause emer-
Accepted 20 June 2014
gency situation from thermal limit consideration. Line flow constraints have not only considered for next
Available online 22 July 2014
predicted interval but in present base case loading conditions also. Optimum load shed at the selected
buses have been obtained for 30-bus, 39-bus standard test systems. Further another technique modified
Keywords:
version of bare bones particle swarm optimization known as (BBExp) has been used to validate the algo-
Load shed
Power system security
rithm for load shedding.
Sensitivity Ó 2014 Elsevier Ltd. All rights reserved.
Teaching learning based optimization
(TLBO)
Modified version of bare bones particle
swarm optimization (BBExp)
Line overloads alleviation

1. Introduction an algorithm for alleviating line overloads by rescheduling genera-

tion and load shedding. Shah and Shahidehpour [5] presented an
Load shedding is required when the system enters into a state algorithm for load shedding based on heuristic rules which are
of non-correctable emergency owing to disturbances on the sys- based on dc power flow model. A load shedding technique that
tem and it has been discussed in Ref. [1]. Non-correctable emer- maximizes the system reactive security margin has been devel-
gency may be encountered owing to (i) voltage limit violations, oped by Berg and Sharaf [6], which is based on non linear optimi-
(ii) line flow limit violations (thermal limit violations), (iii) voltage zation for mutation. Tuan et al. [7] described a load shedding
stability limit and (iv) frequency drop. algorithm which is base on the sensitivity of a voltage stability
In non-correctable emergency state the above mentioned viola- index with respect to load shed. Bijwe et al. [8] used Linear Pro-
tions could not be corrected by rescheduling the usual real/reactive gramming to develop anticipatory load shedding scheme from
power control variables. Hence it becomes necessary to carry out voltage stability consideration view point. Jung et al. [9] used multi
load shedding as one of the key options in power system restora- agent system for development of a new defense system for assess-
tion at selected buses. Load shedding should be distributed at ment of power system vulnerability and perform self healing cor-
selected buses so that inconvenience caused to consumers should rective and preventive control action including load shedding. A
be kept minimum and should not heavily penalize to the consum- linear programming based optimum load shedding algorithm for
ers in a localized area. improving load margin using load margin sensitivity was devel-
Probably Hajdu et al. [2] were the first to develop an algorithm oped by Echavarren et al. [10]. A load shed algorithm accounting
for load shedding using Newton Raphson method and Kuhn-Tucker load dynamics to prevent voltage collapse was developed by
theorem for alleviating line overloads. Palaniswamy et al. [3] pre- Chattopadhyay and Chakrabarti [11]. Pande and Arya [12] used
sented a technique for load shedding accounting load characteris- Hopefield neural network based optimization model for load shed-
tics and generator control effects. Medicherla et al. [4] developed ding to alleviate line overloads. Amraee et al. [13] used static
voltage stability margin as severity index along with its sensitivity
to determine optimum load sheds at selected buses. Girgis and
⇑ Corresponding author.
Mathure [14] utilized rate of change of frequency and voltage vari-
E-mail addresses: [email protected] (L.D. Arya), [email protected]
(A. Koshti). ations for load shedding. Fu and Wang [15] developed a scheme for

http://dx.doi.org/10.1016/j.ijepes.2014.06.066
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 863

Nomenclature

fi flow in ith line xj-min and xj-max lower and upper bounds on jth control variable
fi limit on flow of ith line NP number of population vector
SOL set of overloaded lines NC total number of control variable or decision variable
‘I’ severity index based on overloading NBOL total number of buses selected for load shed
@I
SIp ¼ @P p
sensitivity of severity index with respect to load shed at ui, Ui, rand uniformly distributed random digit between 0 and 1
pth bus ðkÞ
X best best solution vector which gives least value of objective
aip sensitivity of MW-line flow for ith line with respect to function and feasible in kth generation
load shed at pth bus X mean solution vector
J1, J2, J3 and J4 sub jacobian matrices X modified ith modified solution vector obtained in teacher phase
i
S1, S2, S3 and S4 sensitivity sub matrices ðkÞ
Xi ith solution vector in kth generation
DPp load shed at pth bus
SVkp change in kth bus voltage with load shed at pth bus TF teaching factor
SVmp change in mth bus voltage with load shed at pth bus X new
i ith new vector obtained in learner phase
SDkp change in kth bus phase angle with load shed at pth bus kmax maximum number of generation
SDmp change in mth bus phase angle with load shed at pth bus ðkÞ
Pbesti ith particle best position
J objective function (total load shed at selected buses) ðkÞ
Gbest global best amongst all particle
LSB set of selected buses for load shed
ðkÞ ðkÞ
DP p and DP p upper and lower limit of load shed at pth bus l ; r
ij ij
mean and absolute difference of Pbest and Gbest in jth
NL number of lines decision variable of ith particle at kth iteration respec-
fib ith line flow in existing loading condition tively
fip ith line flow under predicted load condition of the load ðkÞ ðkÞ
Nðlij ; rij Þ Normally distributed random variate of mean lðkÞ
ij
shed ðkÞ
and standard deviation as rij
V i and V i lower and upper bounds on load bus voltage magni-
NR total number of independent runs
tudes
Jmin,j minimum value of objective function obtained in the jth
V bi ith load bus voltage under present interval condition of
run
the load shed
J average value of Jmin,j(j = 1,. . . NR)
V pi ith load bus voltage magnitudes after load shed under
predicted next interval condition
r standard deviation of observed minimum value of Jmin
in total independent runs
NG number of generator buses
NB total number of buses
c confidence coefficient
F(,) power flow equations
a constant decided by c
s standard deviation of mean
V0, d0 bus voltages and phase angles in base case condition L length of confidence interval
under pre-load shed condition cv coefficient of variation
Vb, db bus voltages and phase angles in base case condition nb number of better fitness values than mean out of total
after load shed number of runs
Vp, dp bus voltages and phase angles under predicted load con- foc frequency of convergence
dition without load shed ðkÞ
J min minimum value of objective function in kth iteration for
c c
V ,d bus voltages and phase angles under predicted next the best run
interval condition after load shed qp, qq violations in inequality constraint
gl(X) inequality constraints
X(0) initial population

obtaining load shed in emergency conditions for non convergence methods which are the Analytical Hierarchy Process (AHP) and
of power flow program. An adaptive under voltage load shedding the Technique for Order Preference by Similarity to Ideal Solution
scheme using model predictive control to provide protection (TOPSIS) methods. The objective of this paper is to extract the pos-
against voltage instability was developed by Amraee et al. [16]. sible unnecessary load and determine the ranking of the load
Arya et al. [17] used differential evolution to develop an optimal according to their importance level. Arief et al. [23] implemented
load shedding algorithm which provides load shed at selected a new method of under voltage load shedding in a power system
buses in anticipation for the next interval predicted loading condi- incorporating the use of wind generators to maintain voltage sta-
tions based on the minimum Eigen value load flow jacobian. Gu bility following a severe disturbance. The proposed technique in
et al. [18] recently addressed a multi-stage under frequency load this research involves an iterative algorithm based on trajectory
shedding (UFLS) approach to restore frequency of islanded micro- sensitivity analysis to solve the load shedding problem. Sun et al.
grid. Sigrist et al. [19] presented schemes for under frequency load [24] developed a flexible load shedding strategy considering real
shedding for small isolated power systems. Karimi et al. [20] devel- time dynamic thermal line rating (RT-DTLR) technology and imple-
oped an UFLS scheme for isolated distribution network incorpo- mented in operations to improve transmission line capacity, so
rated with mini hydro plant. Optimum load shedding during that congestion costs and/or risk of load shedding can be reduced.
contingency situations is one of the most important issues in It is observed from literature review that objectives of load
power system security analysis. Hooshmand and Moazzami [21] shedding are (i) line overload alleviation, (ii) alleviation of voltage
have presented a fast and optimal adaptive load shedding method, violations, (iii) to maintain desired level of voltage stability margin
for isolated power system using Artificial Neural Networks (ANNs). and (iv) to maintain frequency deviations within limit. Moreover
This method has been tested on the New-England power system. from literature survey it is inferred that anticipatory load shedding
Goh et al. [22] implemented load shedding application toward a for line overload alleviation has not been attempted hence the
large pulp mill electrical system using the multi-decision making objective of this paper is to develop an algorithm for optimum load
864 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

shed at selected buses so as to alleviate the overloads in the pre- @V k

aip ¼ ½2g i V k  g i V m cosðdk  dm Þ þ bi V m sinðdk  dm Þ
dicted load for next interval. The load shed determination is in @Pp
present operating condition such that no violation of thermal load- @V m
ing will occur in next interval. If overloading is going to be exces- þ ½bi V k sinðdk  dm Þ  g i V k cosðdk  dm Þ
@Pp
sive in next interval it may be implemented in present conditions.  
@dk @dm
Since excessive overloading may cause cascaded outages of trans- þ ½bi V k V m cosðdk  dm Þ þ g i V k V m sinðdk  dm Þ 
mission lines and a black out situation may results. Recently devel- @Pp @Pp
oped TLBO is employed for the optimum load shed at selected ð4Þ
buses as this technique does not require any algorithm-specific
parameters. It requires only the control of the common control @V k @V m @dkm
aip ¼ C 1i þ C 2i þ C 3i ð5Þ
parameters such as population size and number of generations @Pp @Pp @Pp
proves effective in optimization. Moreover algorithm is also vali-
The partial derivatives @V k
; @V m
and @dkm
are evaluated from
dated by modified version of bare bones particle swarm optimiza- @Pp @Pp @P p
incremental power flow equations at solution point for the pre-
tion (BBExp).
dicted load conditions are given as
The organization of the paper is as follows: Section 2 presents
a technique for selection of buses for load shed. Section 3 deals     
J1 J2 Dd DP
with problem formulation. Section 4 gives implementation of ¼ ð6Þ
J3 J4 DV DQ
algorithm for load shedding using TLBO. Section 5 gives a compu-
tational algorithm for optimum load shed using BBExp. Section 6 Incremental change in power angles and bus voltages are
gives results for 30-bus and 39-bus test systems. Section 7 pre- obtained from (6) as follows:
sents conclusions. Apart from this Appendices A and B have been     
Dd S1 S2 DP
added for an overview of TLBO and BBExp. Further to handle ¼ ð7Þ
inequality constraints Appendix C has been presented. Moreover DV S3 S4 DQ
30 bus and 39 bus standard test system data is presented in S1, S2, S3 and S4 are sensitivity sub matrices.
Appendices D and E. Eqs. (2)–(7) are required to evaluate sensitivity of severity
index. As Eq. (5) shows the dependency of aip on @V k
@Pp
; @V
@Pp
m
and @d km
@P p
2. Selection of buses for load shed thus it is required to be evaluated using Eqs. (6) and (7).
Assuming constant power factor load operation, load shed DPp
A severity index based on line-overloading condition is defined at pth bus is associated with reactive power load shed as
as follows for the predicted loading conditions as well as for pres-
ent conditions (if overload exist under existing condition) DQ p ¼ bp DP p ð8Þ

1X 2 where bp = tan Up
I¼ ðfi =fi Þ ð1Þ Up is power factor angle of bus p. Change in kth and mth bus
2 i2SOL
voltage and power angles are given as follows:
fi flow in ith line
fi limit on flow of ith line DV k ¼ ðS3kp  DP p þ S4kp  DQ p Þ
SOL set of overloaded lines Putting DQp from (8)

Summation in Eq. (1) is carried out for overloaded lines. Thus ‘I’ DV k ¼ ðS3kp þ S4kp  bp ÞDPp
represents the severity of overloading. Load shed at any particular @V k
The sensitivity @Pp
is given as
bus causes change in line flows and overloading of some of lines
and therefore all those overloaded lines results into change in @V k
¼ SV kp ¼ S3kp þ bp  S4kp ð9Þ
the sensitivity of severity index with respect to load shed at pth @P p
bus and it is given as follows: @V m @dkm
Similarly @P p
; @Pp
are written as follows:
@I X  1 @fi X 
¼ SIP ¼ fi =fi   ¼ fi =fi 2  aip ð2Þ @V m
@Pp i2SOL f i @P p i2SOL ¼ SV mp ¼ S3mp þ bp  S4mp ð10Þ
@Pp
@fi
where aip ¼ @P p
aip shows the sensitivity of ith line flow with respect to load @dk
¼ SDkp ¼ S1kp þ bp  S2kp ð11Þ
shed at pth bus. @Pp
It is to be noted that the sensitivity ‘aip’ depends on complex
load shed at pth bus. The expression for MW-line flow for ith line @dm
connecting kth and mth bus is written as follows: ¼ SDmp ¼ S1mp þ bp  S2mp ð12Þ
@P p
fi ¼ g i V 2k  g i V k V m cosðdk  dm Þ þ bi V k V m sinðdk  dm Þ ð3Þ Putting Eqs. (9)–(12) in relation (5) expression for aip is written
as follows:
Sensitivity aip is written as follows:
aip ¼ C 1i SV kp þ C 2i SV mp þ C 3i ðSDkp  SDmp Þ ð13Þ
@fi @V k @V k @V k
¼ 2g i V k  g i V m cosðdk  dm Þ þ bi V m sinðdk  dm Þ
@Pp @Pp @Pp @Pp Thus using relation (13) sensitivity of severity index (I) with
@V m @V m respect to load shed at pth bus is evaluated. Thus sensitivity SIP
þ bi V k sinðdk  dm Þ  g i V k cosðdk  dm Þ is evaluated under predicted load conditions and buses are identi-
@Pp @Pp
  fied based on highest and decreasing order of SIP. Higher is the
@dk @dm
þ ½bi V k V m cosðdk  dm Þ þ g i V k V m sinðdk  dm Þ  magnitude of the derived sensitivities, lower will be value of load
@Pp @P p
shed. Hence a ranking with decreasing order of magnitude of SIP
After algebraic manipulation aip is written as follows: is prepared and few of top ranked buses are selected for load shed.
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 865

3. Problem formulation for load shedding to alleviate line FðV b ; db Þ ¼ 0 ð21Þ

overloads
FðV p ; dp Þ ¼ 0 ð22Þ
It is stressed here that the load shed formulation is with respect
to next interval predicted load condition such that all operating FðV c ; dc Þ ¼ 0 ð23Þ
constraints are satisfied in present as well as predicted condition.
The load shed should take place at selected buses and that for min- where F(,) represents power flow equations. Eqs. (20)–(22) rep-
imum. Hence the objective is to minimize total load shed as resent load flow solutions in base case pre-load shedding condi-
follows: tions (V0, d0), base case post load shedding (Vb, db) and predicted
X next interval load flow solution (Vp, dp)
J¼ DP p ð14Þ
V0, d0 bus voltages and phase angles in base case condition
p2LSB
under pre-load shed condition
LSB denotes the set of selected buses for load shed. Vb, db bus voltages and phase angles in base case condition after
The objective function (14) represents total load shed at load shed
selected buses. From the economic point of view this must be min- Vp, dp bus voltages and phase angles under predicted load con-
imized. This minimization will further limit to loss of revenue and dition without load shed
inconvenience to a large number of customers. But this minimiza- Vc, dc bus voltages and phase angles under predicted next inter-
tion of total load shed is subject to following constraint. val condition after load shed.

(i) Inequality constraint on load shed at any pth bus 4. Implementation of algorithm for load shedding using TLBO

DP p  DP p  DP p ð15Þ A ranking is prepared based on the value of decreasing order of

magnitude of sensitivity of severity index SIp with respect to load
The above equation clarifies the load shed at any particular bus
shed at pth bus under predicted load conditions. Thus buses are
will lie between the upper and lower limit based on the criticalness
identified for load shed based on SIp.
of buses in view of load shed philosophy at a particular bus.
An overview of TLBO has been discussed in Appendix A. A com-
DPp load shed at pth bus
putational algorithm is developed for optimum load shed at
DP p and DPp are minimum and maximum limits on load shed at
selected buses based on severity index using TLBO. Computational
pth bus respectively
algorithm to solve the formulated problem is presented as follows:
(ii) Inequality constraints on line flows must be satisfied in cur-
Step-(1): An initial population of size ‘NP’ in feasible space is
rent loading conditions as well as in next interval predicted
generated using (A.2).
loading conditions as follows:
T
fib  fi
ð0Þ
ð16Þ Xi ¼ ½DPð0Þ ð0Þ ð0Þ
pi1 ; DP pi2 ; ::; DP piNBOL  ; fori ¼ 1; . . . ; NP ð24Þ

fic 6 fi ð17Þ where NBOL represents total number of buses selected for load
shed
i ¼ 1; 2; . . . ; NL lines Step-(2): Set iteration count k = 1.
ð0Þ
Step-(3): Select best solution vector X best in feasible space as
fi ith line flow limit explained in (A.3) in view of least value of objective function
fib line flow in existing loading condition using relation (14) e.g. optimum load shed at selected buses.
fic ith line flow under predicted load condition of the load shed. Step-(4): Evaluate mean vector for load shed X ðkÞ at selected
buses using relation (A.4).
(iii) Since load shed at selected buses will cause changes in load Step-(5): Obtain modified population X modified in teacher phase
i
bus voltages and hence after load shed the bus voltages using relation (A.6) for current generation.
should be within limits as follows: Step-(6): In case of violation of limits on either side on decision
variables i.e. load shed at selected buses given by relation (15),
V i  V bi  V i ð18Þ it is set to its limiting values.
Step-(7): If X modified
i is found better than current generation pop-
ðkÞ ðkÞ
V i  V ci  V i ð19Þ ulation X i than it replaced by X modified
i otherwise X i is retained
at the end of teaching phase.
i ¼ NG þ 1; . . . ; NB Step-(8): A new solution vector X new i is obtained for all i = 1,. . ., NP
with random selection of vector Xj using relation (A.7) and (A.8).
V i and V i are lower and upper bounds on load bus voltage Step-(9): In case of violation of limits on decision variables it is
magnitudes. set to its limiting values.
V bi ith load bus voltage under present interval condition of the Step-(10): If X new
ðkÞ ðkÞ
is better than X i ; X i is replaced by X new ;
i i
load shed. ðkÞ
otherwise X i is retained in new population.
V ci ith load bus voltage magnitudes after load shed under pre- Step-(11): Evaluate optimum load shed up to current iteration
dicted next interval condition. and increase iteration k = k + 1.
Step-(12): Decision regarding better of two solutions is consid-
(iv) Load flow equations must satisfied under base condition as ered after Step-(6) and Step-(9) according to handling of
well as under predicted next interval condition given as inequality constraints given in (C.1)-(C.4).
follows: Step-(13): Repeat from step-3 till a maximum number of itera-
tions are executed say kmax = 500.
FðV 0 ; d0 Þ ¼ 0 ð20Þ
For statistical inference above sequence is repeated 20 times.
866 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

5. Computational algorithm for optimum load shed using Table 2

Limits on optimum load shed for IEEE-30 bus system.
modified bare bones particle swarm optimization (BBExp)
Sr. no. Bus number (p) Optimum load shed (DPp) p.u. DQ p
b¼ DPp
BBExp is used for optimization of load shed at selected buses Maximum Minimum
which are selected based on sensitivity of severity index. An over-
1 2 0.20 0 0.58525
view of BBExp has been discussed in Appendix B. A computational 2 5 0.30 0 0.2017
algorithm using BBExp to solve the formulated problem is pre- 3 21 0.15 0 0.64
sented as follows: 4 10 0.05 0 0.34483
5 19 0.09 0 0.35789

Step-1: Generate initial population of size ‘NP’ using (B.6) given

as follows:
6.1. Results for 30-bus test system using TLBO and BBExp

ð0Þ ð0Þ ð0Þ

Sð0Þ ¼ ½X 1 ; X 2 ; . . . X NP  ð25Þ The buses for load shed are selected based on sensitivity of
severity index for 30-bus test system. The top ranked buses order
is obtained as 2, 5, 21 10 and 19. Their sensitivities are presented
ð0Þ T in Table 1.
Xi ¼ ½DPð0Þ ð0Þ ð0Þ
pi1 ; DP pi2 ; ::; DP piNBOL  fori ¼ 1; . . . ; NP ð26Þ
Loading in base case and predicted interval without load shed
Each DPpi is randomly generated in the range governed by have been presented in Appendix D.2. Implemented algorithm
inequality constraints (15). These generated samples of load for anticipatory load shedding for present as well as predicted
shed must be selected in feasible space. Obtain objective func- interval have been used and computationally efficient technique
tion for each particle of the population, using power flow TLBO is used to obtain optimal load shed at selected buses. The
program. load shed at selected buses are optimized in the given limits given
in relation (15) mentioned in Table 2. Moreover line flows and volt-
Step-2: Set iteration count k = 1. age limits inequalities as per relation (16)–(19) are also considered
ðkÞ ðkÞ
Step-3: Obtain Pbesti and Gbest based on objective function as for existing and predicted interval for optimization mentioned in
given in relation (B.2) and (B.3) respectively. Appendix D.1 and D.2. The power factor at respective buses are
Step-4: Obtain modified particle using relations (B.7)–(B.10). kept constant during the load shed optimization by keeping con-
stant ratio of reactive power load shed to the real power load shed
When the resulting positions of each individual does not satisfy i.e. (b) and shown in Table 2.
constraints on decision variables, in such situations it is set to In TLBO algorithm NP = 15 decision variable vectors are consid-
limiting values. ered for optimization. Each vector consists of the decision variables
equals to load shed at selected buses. An initial population is gen-
Step-5: Decision regarding better of two solutions is considered erated in the range governed by inequality (15) for load shed at
according to handling of inequality constraints given in (C.1)- buses. The limits on load shed at selected buses are given in Table 2.
(C.4). Twenty independent runs have been carried out with maximum
Step-6: Objective function is evaluated for updated vector and number of generation specified as kmax = 500 as explained in Sec-
ðkþ1Þ ðkþ1Þ
increase k = k + 1.P besti and Gbest are obtained using (B.11) tion 4. No significant changes have been observed after 88 genera-
and (B.12) and procedure is repeated from Step-3 for a maxi- tions for best run and time for convergence is obtained as
mum number of generation specified (kmax = 500) or The proce- 20.2350 s. for 30-bus system. Optimal load shed DPp (p.u.) with
dure may even terminated if there is no improvement in satisfying all operating constraints are obtained as 0.016527,
objective function is observed for a specified number of 0.048211, 0.017382, 0.029803 and 0.089986 at buses 2, 5, 21, 10
iterations. and 19 respectively for best run. The total minimum optimum load
P
shed DPp (p.u.) is obtained as 0.201909. Moreover this algorithm
For statistical inference above sequence is repeated 20 times. is also validated using BBExp with NP = 15. The total minimum
optimum load shed obtained using BBExp as 0.202142 p.u. A com-
parison for optimum load shed using TLBO and BBExp is shown in
Table 3. Twenty independent runs have been carried out for BBExp
6. Results and discussions
with maximum number of generation specified as kmax = 500 as
explained in Section 5. No significant changes have been observed
The developed algorithm has been implemented on the stan-
after 95 generations. Time for convergence is obtained as 25.
dard 30-bus and 39-bus test systems. The system data is given in
5082 s. for 30-bus system for best run. The results obtained using
Appendices D and E. The results have been obtained on Intel Pen-
both the techniques are observed in close agreement. A graphical
tium Dual CPU processor, 2 GB RAM, 2.0 GHz configuration and
plot for best run for optimum load shed using TLBO and BBExp is
presented for 30-bus and 39-bus test systems. ðkÞ
plotted J min with number of generation for 30-bus test system as
illustrated in Fig. 1.
Bus voltage magnitudes for base case and predicted interval
Table 1
with and without load shed are presented as shown in Table 4
Top ranked buses order based on sensitivity of severity index with respect to load
shed for IEEE-30 bus system.
for 30-bus system. Moreover line flows are also compared for all
  conditions discussed above and given in Table 5. It is observed
Sr. no. Bus number (p)  104
Sensitivity @I
@Pp from this table that some of the lines 1, 8 and 15 are found over-
1 2 2557.6
loaded in predicted interval condition violating the line-flow limits
2 5 2551.68 as given in Appendix D.1. This lines are (⁄) marked given in Table 5
3 21 1672.64 for pre load shed predicted interval case line flows. These lines
4 10 1671.04 have been relieved with planning of minimum load shed at
5 19 1670.56
selected buses. This is reflected in predicted interval line flows case
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 867

Table 3
Optimum load shed obtained of best run for IEEE-30 bus system using TLBO and BBEXp.

Sr. no. Bus number (p) Optimum load shed using TLBO (DPp) p.u. Optimum load shed using BBExp (DPp) p.u.
1 2 0.016527 0.017234
2 5 0.048211 0.049510
3 21 0.017382 0.027382
4 10 0.029803 0.028030
5 19 0.089986 0.079986
P
Total optimum load shed (p.u.) Jmin = (DPp) 0.201909 0.202142

Total
Load shed
(k)
J min
(p.u.)

Number of Generation (k)

ðkÞ
Fig. 1. Graphical plots for the best run for minimum value of load shed Jmin v/s number of generation for IEEE-30 bus system.

Table 4
Comparison of Bus voltage magnitudes for IEEE-30 bus system using TLBO and BBExp.

Bus Base case voltage magnitudes Predicted interval voltage magnitudes

number
Pre load shed After load shed using TLBO After load shed using BBExp Pre load shed After load shed using TLBO After load shed using
(V0) p.u. (Vb) p.u. (Vb) p.u. (Vp) p.u. (Vc) p.u. BBExp (Vc)
1 1.06 1.06 1.06 1.06 1.06 1.06
2 1.043 1.043 1.043 1.043 1.043 1.043
3 1.0278 1.0292 1.037 1.0256 1.0272 1.0349
4 1.0202 1.022 1.0314 1.0175 1.0195 1.0288
5 1.01 1.01 1.01 1.01 1.01 1.01
6 1.0162 1.0178 1.0218 1.014 1.0157 1.0197
7 1.005 1.0059 1.0082 1.0025 1.0035 1.0059
8 1.01 1.01 1.01 1.01 1.01 1.01
9 1.0511 1.055 1.0397 1.0472 1.0514 1.0361
10 1.0426 1.0497 1.0366 1.0361 1.0434 1.0303
11 1.082 1.082 1.082 1.082 1.082 1.082
12 1.0601 1.0628 1.039 1.0561 1.059 1.0352
13 1.071 1.071 1.071 1.071 1.071 1.071
14 1.0434 1.0477 1.0252 1.0372 1.0418 1.0192
15 1.0372 1.0438 1.0227 1.0303 1.0372 1.016
16 1.0445 1.0492 1.0297 1.0385 1.0434 1.0239
17 1.0372 1.0437 1.0285 1.0301 1.0368 1.0217
18 1.0256 1.0397 1.0205 1.0172 1.0316 1.0124
19 1.0223 1.0407 1.0227 1.0134 1.0322 1.0141
20 1.0265 1.042 1.0252 1.018 1.034 1.0171
21 1.029 1.0368 1.0244 1.0207 1.0289 1.0165
22 1.0296 1.0372 1.0248 1.0214 1.0294 1.0169
23 1.0244 1.0309 1.0139 1.0157 1.0225 1.0055
24 1.0167 1.023 1.0116 1.0069 1.0135 1.0021
25 1.0129 1.0166 1.0146 1.0039 1.0079 1.0058
26 0.99337 0.99714 0.99503 0.98185 0.9859 0.98377
27 1.0204 1.0225 1.0263 1.013 1.0154 1.0191
28 1.0132 1.0147 1.0165 1.0105 1.0122 1.0139
29 0.99835 1.0006 1.0044 0.98811 0.99057 0.99441
30 0.98562 0.98786 0.99174 0.97373 0.97622 0.98013
868 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

Table 5
Comparison of line flows for IEEE-30 bus system using TLBO and BBExp.

Line Base case line flows Predicted interval line flows

no.
Pre load shed After load shed using TLBO After load shed using BBExp Pre load shed After load shed using TLBO After load shed using
ðfi0 Þ p.u. ðfib Þ p.u. ðfib Þ p.u. ðfip Þ p.u. ðfic Þ p.u. BBExp ðfic Þ p.u.

1 0.5143 0.40658 0.40532 0.7567⁄ 0.61316 0.61083

2 0.34238 0.28646 0.2723 0.44849 0.38267 0.37733
3 0.25602 0.22377 0.21432 0.30643 0.27226 0.267
4 0.31367 0.25907 0.24453 0.4107 0.34937 0.3431
5 0.47981 0.43424 0.43297 0.58293 0.53597 0.53437
6 0.32243 0.27913 0.27667 0.39588 0.3501 0.34998
7 0.30902 0.26408 0.32073 0.40162 0.35444 0.38997
8 0.1076 0.0987 0.084831 0.14024⁄ 0.13096 0.11649
9 0.32529 0.32308 0.32604 0.37158 0.36917 0.37135
10 0.1512 0.18928 0.28405 0.10145 0.13731 0.22854
11 0.08397 0.07343 0.16834 0.10632 0.07902 0.16652
12 0.09337 0.06547 0.083762 0.11423 0.08568 0.10551
13 0.34439 0.33477 0.37688 0.35469 0.34365 0.38869
14 0.36843 0.31607 0.3154 0.40838 0.35471 0.35302
15 0.21208 0.1748 0.11775 0.25926⁄ 0.21534 0.15865
16 0.4108 0.40652 0.47439 0.41879 0.41265 0.49063
17 0.09708 0.08786 0.084154 0.10795 0.09863 0.094698
18 0.23803 0.20248 0.19139 0.2637 0.22776 0.21543
19 0.11115 0.09868 0.088266 0.12219 0.10959 0.097682
20 0.02514 0.0162 0.013565 0.02759 0.01859 0.015374
21 0.06747 0.05561 0.050811 0.07302 0.06103 0.054079
22 0.08112 0.03727 0.038151 0.08949 0.04523 0.044964
23 0.0439 0.00901 0.018216 0.04782 0.00912 0.017086
24 0.06943 0.01887 0.035177 0.07795 0.02666 0.04284
25 0.09599 0.04307 0.058878 0.10782 0.05451 0.070319
26 0.06293 0.07281 0.093948 0.06977 0.07999 0.10101
27 0.20904 0.2007 0.18926 0.23275 0.22437 0.21288
28 0.10015 0.09808 0.092549 0.11142 0.10932 0.10376
29 0.02404 0.01555 0.016701 0.02674 0.01741 0.018296
30 0.08277 0.08458 0.072207 0.09061 0.09242 0.07889
31 0.07786 0.08901 0.084923 0.08568 0.0968 0.092598
32 0.04309 0.04493 0.036943 0.04599 0.04782 0.038008
33 0.01845 0.03162 0.029483 0.01627 0.02943 0.028207
34 0.04698 0.04697 0.046973 0.05246 0.05245 0.052452
35 0.0328 0.02593 0.051194 0.04053 0.03244 0.057759
36 0.18323 0.17037 0.19191 0.21064 0.1976 0.21971
37 0.07075 0.07074 0.070723 0.07914 0.07913 0.079104
38 0.08041 0.08039 0.080373 0.08997 0.08995 0.089925
39 0.04135 0.04134 0.041338 0.04614 0.04614 0.04613
40 0.06029 0.06444 0.073678 0.04989 0.05358 0.062469
41 0.15477 0.14849 0.17748 0.17584 0.16892 0.19639

Table 6
Statistics for optimum load shed using TLBO and BBExp for IEEE-30 bus system.

Sr. Statistical parameter obtained for optimum load shed (p.u.) Statistical parameter value obtained using Statistical parameter value obtained using
no. TLBO BBExp
Number of runs NR = 20
1 Average value J 0.217268 0.219256
2 Worst minimum value Max[Jmin,1, Jmin,2,. . ., Jmin,NR] 0.229403 0.229956
3 Best fitness value Min[Jmin,1, Jmin,2, . . ., Jmin,NR] 0.201909 0.202142
4 Standard deviation (r) 0.00785032 0.0081023
5 Standard deviation of mean s ¼ prffiffiffiffiffi
NR
0.00175538 0.001811729
6 Confidence interval with, CONF 0:95 ½J  c:s; J þ c:s for [0.21380991, 0.22072609] [0.215686893, 0.222825107]
c = 0.95 a = 1.97
7 Length of Interval L 0.00691618 0.007138214
8 Coefficient of variation (cv) 0.036131966 0.036953607
 nb

9 Frequency of convergence foc ¼ NR 0.70 0.60
10 Median value [Jmin,1, . . ., Jmin,NR] 0.217275 0.2218130

after load shed. The results for optimum load shed using TLBO and TLBO as 0.217268, 0.201909 and 0.229403, whereas median value
BBExp is provided for best run. is obtained as 0.217275. Whereas using BBExp average value, best
Various statistics has been presented with 20 independent runs fitness value and worst minimum value of load shed (p.u.)
for optimum load shed using TLBO and BBExp for 30-bus test sys- obtained as 0.219256, 0.202142 and 0.229956. Median value is
tem given in Table 6. This table shows average value, best fitness obtained as 0.2218130. This table also illustrated the standard
value and worst minimum value of load shed (p.u.) obtained using deviation (r) of samples obtained for load shed in p.u. obtained
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 869

Table 7 load shed at selected buses have been optimized in the given limits
Top ranked buses order based on sensitivity of severity index with respect to load specified in Table 8 according to the relation (15). Moreover line
shed for IEEE-39 bus system.
flows and voltage limits inequalities as per relations (16)–(19)
 
Sr. no. Bus number Sensitivity @I
@Pp  104 are also considered for existing and predicted interval for optimi-
zation mentioned in Appendices E.1 and E.2. The power factor at
1 29 284.96
respective buses are kept constant during the load shed optimiza-
2 28 272.48
3 25 242.88 tion by keeping constant ratio of reactive power load shed to the
4 26 235.68 real power load shed i.e. (b) and shown in Table 8.
5 18 223.04 In TLBO algorithm NP = 15 decision variable vectors are consid-
ered for optimization. Each vector consists of the decision variables
equals to load shed at selected buses. An initial population is gen-
Table 8 erated in the range governed by inequality (15) for load shed at
Limits on optimum load shed for IEEE-39 bus system. buses. The limits on load shed at selected buses are given in Table 8.
Sr. no. Bus number (p) Optimum load shed (DPp) p.u. DQ p Twenty independent runs have been carried out with maximum
b¼ DPp
number of generation specified as kmax = 500 as explained in Sec-
Maximum Minimum
tion 4. No significant changes have been observed after 95 genera-
1 29 0.5 0 0.094885 tions for best run and time for convergence is obtained as 25.761 s.
2 28 0.5 0 0.133982
3 25 0.5 0 0.210715
for 39-bus system. Optimal load shed DPp (p.u.) satisfying all oper-
4 26 0.5 0 0.122303 ating constraints are obtained as 0.15611, 0.19461, 0.029178,
5 18 0.5 0 0.189874 0.14596 and 0.16355 at buses 29, 28, 25, 26 and 18 respectively
P
for best run. The total minimum optimum load shed DPp (p.u.)
is obtained as 0.68941. Moreover this algorithm is also validated
using TLBO as 0.00785032 with confidence interval for c = 0.95. using BBExp with NP = 15. The total minimum optimum load shed
Whereas for BBExp it is obtained as 0.0081023. Other statistics is obtained using BBExp as 0.698812 p.u. A comparison for optimum
also demonstrated in Table 6 and it is observed that frequency of load shed using TLBO and BBExp is shown in Table 9. Twenty inde-
convergence is coming more i.e. 0.7 using TLBO then 0.6 in BBExp. pendent runs have been carried out for BBExp with maximum
number of generation specified as kmax = 500 as explained in Sec-
tion 5. No significant changes have been observed after 105 gener-
6.2. Results for 39-bus test system using TLBO and BBExp ations. Time for convergence is obtained as 35.563 s. for 39-bus
system for best run. The results obtained using both the techniques
The buses for load shed are selected based on sensitivity of are observed in close agreement. A graphical plot for best run for
ðkÞ
severity index for 39-bus test system. The top ranked buses order optimum load shed using TLBO and BBExp is plotted J min with num-
have been obtained as 29, 28, 25, 26, and 18. The sensitivities of ber of generation for 39-bus test system as illustrated in Fig. 2.
buses have been presented for load shed in Table 7. Bus voltage magnitudes for base case and predicted interval
Base case and predicted interval loading without load shed have with and without load shed are presented as shown in Table 10
been presented in Appendix E.2. TLBO algorithm have been devel- for 39-bus system. Moreover line flows are also compared for all
oped to obtain optimal load shed at selected buses using anticipa- conditions discussed above and given in Table 11. It is observed
tory load shedding for present as well as predicted interval. The from this table that some of the lines 6, 7, 13 and 19 are found

Table 9
Optimum load shed obtained of best run for IEEE-39 bus system using TLBO and BBEXp.

Sr. no. Bus number Optimum load shed using TLBO (DPp) p.u. Optimum load shed using BBExp (DPp) p.u.
1 29 0.15611 0.16128
2 28 0.19461 0.20764
3 25 0.029178 0.01959
4 26 0.14596 0.13979
5 18 0.16355 0.170512
P
Total optimum load shed (p.u.) Jmin = (DPp) 0.68941 0.698812

Total
Load shed
(k)
J min
(p.u.)

Number of Generation (k)

ðkÞ
Fig. 2. Graphical plots for the best run for minimum value of load shed J min v/s number of generation for IEEE-39 bus system.
870 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

Table 10
Comparison of Bus voltage magnitudes for IEEE-39 bus system using TLBO and BBExp.

Bus Base case voltage magnitudes Predicted interval voltage magnitudes

number
Pre load shed After load shed using TLBO After load shed using BBExp Pre load shed After load shed using After load shed using
(V0) p.u. (Vb) p.u. (Vb) p.u. (Vp) p.u. TLBO (Vc) p.u. BBExp (Vc) p.u.
1 1.0653 1.0665 1.0666 1.0594 1.0612 1.0614
2 1.0681 1.0686 1.0686 1.0636 1.0645 1.0646
3 1.0536 1.0544 1.0545 1.047 1.0483 1.0484
4 1.0233 1.0242 1.0243 1.015 1.0163 1.0164
5 1.0199 1.021 1.0211 1.011 1.0126 1.0127
6 1.0214 1.0224 1.0225 1.013 1.0145 1.0146
7 1.0092 1.0105 1.0107 0.99845 1.0004 1.0006
8 1.0074 1.0089 1.009 0.99572 0.99787 0.99807
9 1.033 1.0351 1.0352 1.0212 1.0239 1.0242
10 1.0314 1.0321 1.0321 1.0251 1.0262 1.0263
11 1.027 1.0278 1.0279 1.02 1.0212 1.0213
12 1.014 1.0147 1.0148 1.006 1.0071 1.0073
13 1.0308 1.0315 1.0316 1.0241 1.0252 1.0253
14 1.0327 1.0335 1.0336 1.0256 1.0267 1.0269
15 1.0402 1.0408 1.0409 1.0343 1.0352 1.0353
16 1.0567 1.0573 1.0574 1.0523 1.0531 1.0532
17 1.0638 1.0647 1.0648 1.0584 1.0596 1.0597
18 1.0594 1.0605 1.0606 1.0534 1.0549 1.055
19 1.0597 1.06 1.06 1.0579 1.0582 1.0582
20 0.99538 0.9955 0.99551 0.99373 0.9939 0.99392
21 1.0522 1.0526 1.0527 1.0486 1.0492 1.0492
22 1.0627 1.0629 1.0629 1.0607 1.061 1.061
23 1.0583 1.0585 1.0586 1.0562 1.0565 1.0565
24 1.0622 1.0627 1.0628 1.0583 1.0591 1.0591
25 1.0721 1.0738 1.074 1.066 1.0681 1.0683
26 1.0935 1.0949 1.0952 1.088 1.0897 1.09
27 1.0758 1.0769 1.0771 1.0698 1.0712 1.0714
28 1.0869 1.0883 1.0888 1.0836 1.0852 1.0858
29 1.0765 1.0774 1.0781 1.0742 1.0753 1.076
30 1.0475 1.0475 1.0475 1.0475 1.0475 1.0475
31 0.982 0.982 0.982 0.982 0.982 0.982
32 0.9831 0.9831 0.9831 0.9831 0.9831 0.9831
33 0.9972 0.9972 0.9972 0.9972 0.9972 0.9972
34 1.0123 1.0123 1.0123 1.0123 1.0123 1.0123
35 1.0493 1.0493 1.0493 1.0493 1.0493 1.0493
36 1.0635 1.0635 1.0635 1.0635 1.0635 1.0635
37 1.0278 1.0278 1.0278 1.0278 1.0278 1.0278
38 1.0265 1.0265 1.0265 1.0265 1.0265 1.0265
39 1.03 1.03 1.03 1.03 1.03 1.03

overloaded in predicted interval condition violating the line-flow interval predicted load as well as the present loading condition
limits given in Appendix E.1. This lines are (⁄) marked given in which could cause emergency situation from thermal limit consid-
Table 11 for pre load shed predicted interval case line flows. These eration. This load shed at selected buses have been minimized
lines have been relieved with planning of minimum load shed at using a computationally efficient technique (TLBO) which does
selected buses. This is reflected in predicted interval line flows case not require any algorithm-specific parameters and thus requires
after load shed. The results for optimum load shed using TLBO and less computational efforts. The methodology is also validated using
BBExp is provided for best run. another evolutionary technique BBExp. Constraints such as line
Various statistics has been presented with 20 independent runs flows, voltage limits and power flow equations are considered for
for optimum load shed using TLBO and BBExp for 39-bus test sys- all operating condition for predicted next interval and present
tem given in Table 12. This table shows average value, best fitness loading condition. This developed algorithm has been tested for
value and worst minimum value of load shed (p.u.) obtained as standard 30-bus and 39-bus test systems. The results obtained
0.7009753, 0.689408 and 0.711973, whereas median value is for optimum load shed using both methods are in close agreement
obtained as 0.70129. Whereas using BBExp average value, best fit- but TLBO looks more efficient in view of frequency of convergence
ness value and worst minimum value of load shed (p.u.) obtained and time for convergence for best run. This developed algorithm is
as 0.7198223, 0.698812 and 0.732789. Median value is obtained a promising method shows its utility for power system planners to
as 0.71994. This table also illustrated the standard deviation (r) plan load shed under emergency situation and in view of future
of samples of load shed in p.u. obtained as 0.0070442 with confi- predicted load demand if it comes then may cause excessive over-
dence interval for c = 0.95. Whereas for BBExp it is obtained as heating and therefore require minimum load shed at appropriate
0.0079376. It is observed from Table 12 that frequency of conver- buses so that less inconvenience should occur for consumers.
gence is coming more i.e. 0.75 using TLBO then 0.65 in BBExp.
Appendix A. Teaching learning based optimization (TLBO): an
7. Conclusions overview

This paper presents a preventive load shedding methodology TLBO algorithm has been recently developed by Rao et al. [25]
which determines optimum load shedding at selected buses based which is population based technique. The methodology is based
on sensitivity of severity index. Load shed is based on the next on the philosophy teaching and learning. The technique accounts
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 871

Table 11
Comparison of line flows for IEEE-39 bus system using TLBO and BBExp.

Line Base case line flows Predicted interval line flows

no.
Pre load shed After load shed using TLBO After load shed using BBExp Pre load shed After load shed using TLBO After load shed using BBExp
ðfi0 Þ p.u. ðfib Þ p.u. ðfib Þ p.u. ðfip Þ p.u. ðfic Þ p.u. ðfic Þ p.u.

1 3.61 3.2258 3.1909 4.9067 4.5079 4.4701

2 4.1589 3.8626 3.8362 5.22 4.8812 4.8495
3 6.7993 6.7054 6.6961 7.5431 7.4463 7.4365
4 0.91044 1.2255 1.2535 0.3613 0.59626 0.626
5 2.7182 2.8286 2.8383 2.7236 2.8303 2.8401
6 1.1958 1.0325 1.0187 1.7395⁄ 1.5409 1.5226
7 1.1496 0.97747 0.96189 1.6209⁄ 1.444 1.4272
8 2.3017 2.336 2.3398 2.1608 2.1918 2.1955
9 1.9014 1.8775 1.8754 1.9686 1.9391 1.9365
10 1.4548 1.5101 1.5162 1.5163 1.5006 1.5003
11 2.2668 2.3467 2.3545 2.1847 2.2344 2.2397
12 3.8219 3.9273 3.9373 3.5214 3.6219 3.6319
13 0.90893 0.78028 0.76853 1.4148⁄ 1.2772 1.264
14 5.7745 5.475 5.4472 7.0272 6.7242 6.695
15 5.912 5.643 5.6183 7.0607 6.7727 6.7452
16 3.8646 3.9582 3.9672 3.6137 3.7015 3.7102
17 2.7065 2.6054 2.5958 3.0082 2.9069 2.897
18 2.6692 2.5605 2.5503 2.9883 2.8794 2.8688
19 0.87966 0.80312 0.79701 1.2719⁄ 1.151 1.1394
20 3.5341 3.6575 3.6699 3.3266 3.4371 3.4488
21 1.1788 1.1393 1.1371 0.86929 0.91094 0.91736
22 4.4522 4.4539 4.4541 4.128 4.1297 4.1299
23 3.0227 3.0262 3.0266 2.8217 2.8265 2.8271
24 1.0561 1.0475 1.0466 1.1855 1.1738 1.1725
25 1.0427 1.0789 1.09 0.81621 0.80889 0.81407
26 1.115 1.1794 1.1933 1.0733 1.143 1.1583
27 5.9792 5.9783 5.9782 5.9354 5.9337 5.9335
28 0.84546 0.84493 0.84487 0.88865 0.88793 0.88786
29 3.4414 3.4429 3.4431 3.3637 3.3657 3.3659
30 2.5307 2.2989 2.2712 2.9352 2.6923 2.6616
31 3.5522 3.7582 3.7866 3.6911 3.8989 3.9288
32 1.6262 1.7796 1.8057 1.4868 1.6393 1.6658
33 2.3067 2.4264 2.4495 2.1838 2.3055 2.3295
34 3.6553 3.6729 3.6952 3.5994 3.6179 3.6417
35 0.45084 0.45207 0.45219 0.46926 0.47005 0.47013
36 0.54586 0.54681 0.54692 0.56967 0.56973 0.56976
37 1.0828 1.0468 1.0436 1.3889 1.3356 1.3306
38 6.6473 6.6407 6.6401 6.7139 6.702 6.7009
39 6.3361 6.3351 6.335 6.3461 6.3443 6.3441
40 5.2757 5.2739 5.2737 5.3011 5.2985 5.2982
41 6.6094 6.6065 6.6062 6.6368 6.6323 6.6318
42 5.6205 5.6198 5.6197 5.6284 5.6271 5.627
43 5.5319 5.5502 5.552 5.4762 5.4939 5.4957
44 2.527 2.5229 2.5226 2.5764 2.564 2.5629
45 8.6255 8.6437 8.6567 8.5822 8.6019 8.6152
46 1.8712 1.8721 1.8721 2.1821 2.183 2.1831

Table 12
Statistics for optimum load shed using TLBO and BBExp for IEEE-39 bus system.

Sr. Statistical parameter obtained for optimum load shed (p.u.) Statistical parameter value obtained using Statistical parameter value obtained using
no. TLBO BBExp
Number of runs NR = 20
1 Average value J 0.7009753 0.7198223
2 Worst minimum value Max[Jmin,1, Jmin,2, . . ., Jmin,NR] 0.711973 0.732789
3 Best fitness value Min[Jmin,1, Jmin,2, . . ., Jmin,NR] 0.689408 0.698812
4 Standard deviation (r) 0.0070442 0.0079376
5 Standard deviation of mean s ¼ prffiffiffiffi
NR
ffi 0.0015751 0.0017749
6 Confidence interval with, CONF 0:95 ½J  c:s; J þ c:s for [0.697872, 0.704078] [0.716325, 0.723318]
c = 0.95 a = 1.97
7 Length of Interval L 0.0062060 0.006993
8 Coefficient of variation (cv) 0.010049141 0.0110271660
nb
9 Frequency of convergence foc ¼ NR 0.75 0.65
10 Median value [Jmin,1, . . ., Jmin,NR] 0.70129 0.71994

the influence of a teacher on the output of students (learners) in who shares knowledge with the students. Naturally the quality
the class. The output is considered in terms of grades/marks. of teacher affects the outcome of students. Learning is accom-
Usually the teacher is supposed to be a highly learned person plished using two ways for learner (i) through teacher known as
872 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

Table D.1
Line data.

Line no. From bus no. To bus no. Line parameter Line flow limit
ri (p.u.) xi (p.u.) Half-line charging susceptance (p.u.) Tap ratio (p.u.)
1 1 2 0.0192 0.0575 0.0264 – 0.71802
2 1 3 0.0452 0.1852 0.0204 – 0.4654
3 2 4 0.0570 0.1737 0.0184 – 0.34815
4 3 4 0.0132 0.0379 0.0042 – 0.42549
5 2 5 0.0472 0.1983 0.0209 – 0.67057
6 2 6 0.0581 0.1763 0.0187 – 0.4503
7 4 6 0.0119 0.0414 0.0045 – 0.49524
8 5 7 0.0460 0.1160 0.0102 – 0.13174
9 6 7 0.0267 0.0820 0.0085 – 0.45977
10 6 8 0.0120 0.0420 0.0045 – 0.34237
11 6 9 0 0.2080 0 1.0155 0.23261
12 6 10 0 0.5560 0 0.9629 0.1589
13 9 11 0 0.2080 0 – 0.54628
14 9 10 0 0.1100 0 – 0.51262
15 4 12 0 0.2560 0 1.0129 0.21582
16 12 13 0 0.1400 0 – 0.67906
17 12 14 0.1231 0.2559 0 – 0.12993
18 12 15 0.0662 0.1304 0 – 0.31371
19 12 16 0.0945 0.1987 0 – 0.1408
20 14 15 0.2210 0.1997 0 – 0.03015
21 16 17 0.0824 0.1932 0 – 0.08536
22 15 18 0.1070 0.2185 0 – 0.10623
23 18 19 0.0639 0.1292 0 – 0.05615
24 19 20 0.0340 0.0680 0 – 0.10894
25 10 20 0.0936 0.2090 0 – 0.14616
26 10 17 0.0324 0.0845 0 – 0.11692
27 10 21 0.0348 0.0749 0 – 0.28687
28 10 22 0.0727 0.1499 0 – 0.13648
29 21 22 0.0116 0.0236 0 – 0.03999
30 15 23 0.1000 0.2020 0 – 0.1002
31 22 24 0.1150 0.1790 0 – 0.10056
32 23 24 0.1320 0.2700 0 – 0.05032
33 24 25 0.1885 0.3292 0 – 0.03172
34 25 26 0.2544 0.3800 0 – 0.06577
35 25 27 0.1093 0.2087 0 – 0.08138
36 27 28 0.0 0.3960 0 0.9581 0.28704
37 27 29 0.2198 0.4153 0 – 0.09903
38 27 30 0.3202 0.6027 0 – 0.11254
39 29 30 0.2399 0.4533 0 – 0.05788
40 8 28 0.0636 0.2000 0.0214 – 0.09669
41 6 28 0.0169 0.0599 0.0065 – 0.25586

teacher phase, (ii) interaction between learners known as learner Thus an initial population is generated using (A.2) is repre-
phase. In this algorithm an optimization problem is optimized con- sented in matrix form of size (NP  NC). Where ‘NP’ represents
sidering tuning variables for optimization problem as different number of learners in the class and ‘NC’ number of subjects offered
subjects (grades/marks) and assuming different learners as popula- in the class
tion. The best learner is treated as teacher for all subjects e.g. best
th
solution amongst population based on fitness function. j subject
An optimization problem J() is optimized using TLBO in follow- #
ing sections: 2 ð0Þ ð0Þ ð0Þ ð0Þ
3
x11 x12 . . . x1j . . . x1NC
6 7
6 . 7
6 .. 7
Appendix A.1. Initialization 6 7
6 . 7
6 . 7 i
th
learner
6 . 7
The initial population X(0) is initialized randomly in search 6 ð0Þ 7
X ð0Þ
¼6
6 xi1
ð0Þ
xi2
ð0Þ
. . . xij
ð0Þ
. . . xiNC 7
7
space using following relation. 6 7
6 .. 7
h i 6 . 7
ð0Þ ð0Þ ð0Þ ð0Þ 6 7
Xi ¼ xi1 ; . . . xij . . . xiNC 6 . 7
ðA:1Þ 6 .. 7
4 5
fori ¼ 1; . . . ; NP ð0Þ ð0Þ ð0Þ ð0Þ
xNP1 xNP2 . . . xNPj . . . xNPNC
NPNC
ð0Þ
xij i.e. jth subject grades/marks of ith learner is obtained from uni-
form distribution as follows
Appendix A.2. Teacher phase
ð0Þ
xij ¼ xj-min þ ðxj-max  xj-min Þui ðA:2Þ
The best learner amongst all learners for all subjects is selected
xj-min and xj-max are lower and upper bounds on variable xi-j. ui is a as teacher who is having higher marks/grades in all subjects.
random digit in the range [0, 1]. Teacher trains the other students in the class. The learner with
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 873

Table D.2
Load data.

Bus no. Bus voltage limit Vi (p.u.) Base case load condition (without load shed) Predicted load condition (without load shed)

Vi Vi PD (p.u.) QD (p.u.) PD (p.u.) QD (p.u.)

1 – – 0 0 0 0
2 – – 0.2387 0.1397 0.26583 0.15558
3 0.97 1.1 0.0264 0.0132 0.0294 0.0147
4 0.97 1.1 0.0836 0.0176 0.0931 0.0196
5 – – 1.0362 0.209 1.154 0.23275
6 0.97 1.1 0 0 0 0
7 0.97 1.1 0.2508 0.1199 0.2793 0.13353
8 – – 0.33 0.33 0.3675 0.3675
9 0.97 1.1 0 0 0 0
10 0.97 1.1 0.0638 0.022 0.07105 0.0245
11 – – 0 0 0 0
12 0.97 1.1 0.1232 0.0825 0.1372 0.09188
13 – – 0 0 0 0
14 0.97 1.1 0.0682 0.0176 0.07595 0.0196
15 0.97 1.1 0.0902 0.0275 0.10045 0.03063
16 0.97 1.1 0.0385 0.0198 0.04288 0.02205
17 0.97 1.1 0.099 0.0638 0.11025 0.07105
18 0.97 1.1 0.0352 0.0099 0.0392 0.01103
19 0.97 1.1 0.1045 0.0374 0.11638 0.04165
20 0.97 1.1 0.0242 0.0077 0.02695 0.00858
21 0.97 1.1 0.1925 0.1232 0.21438 0.1372
22 0.97 1.1 0 0 0 0
23 0.97 1.1 0.0352 0.0176 0.0392 0.0196
24 0.97 1.1 0.0957 0.0737 0.10658 0.08208
25 0.97 1.1 0 0 0 0
26 0.97 1.1 0.0385 0.0253 0.04288 0.02818
27 0.97 1.1 0 0 0 0
28 0.97 1.1 0 0 0 0
29 0.97 1.1 0.0264 0.0099 0.0294 0.01103
30 0.97 1.1 0.1166 0.0209 0.12985 0.02328

Table D.3
Generator data.

Generator no. Voltage magnitude (p.u.) Real generation PG (p.u.) Reactive generation QG (p.u.)
Maximum Minimum Maximum Minimum
1 1.060 2.0000 0.5000 1.5000 0.2000
2 1.043 0.8000 0.2000 0.6000 0.2000
5 1.010 0.5000 0.1500 0.6250 0.1500
8 1.010 0.3500 0.1000 0.5000 0.1500
11 1.082 0.3000 0.1000 0.4000 0.1000
13 1.071 0.4000 0.1200 0.4500 0.1500

The teacher puts his efforts to increase the mean result of the
Table D.4
class in the subject and shifting the mean toward his own mean.
Shunt capacitor data.
Thus difference between results of mean for teacher and learner
Bus number p.u. capacity is given as follows:
10 0.19 ðkÞ
24 0.04 Difference mean ¼ randð:Þ½X best  TF  X ðkÞ  ðA:5Þ
Further any ith existing learner’s results (marks/grades) in all
subjects is modified based on the value of Difference_mean and
minimum objective function value is considered as the teacher for updated population is prepared as follows for all learners i.e.
respective generation ‘k’ and given as: i = 1,. . ., NP:
8 modified ðkÞ ðkÞ
9
ðkÞ
X best ¼ arg min JðX i Þ
ðkÞ
ðA:3Þ
>
< Xi ¼ X i þ randð:Þ½X best  TF  X ðkÞ  >
=
TF ¼ 1 if Ui  0:5 ðA:6Þ
>
: >
;
J(  ) is the objective function ¼ 2 if Ui > 0:5
The mean of all marks of the learners for each subject in the
Ui and rand() are random numbers in the range [0, 1].
class for generation ‘k’ is evaluated and thus mean vector X ðkÞ given
TF is known as teaching factor which decides value of the mean
as follows:
to be changed, and its value is either 1 or 2 as explained in relation
h i (A.6).
ðkÞ ðkÞ ðkÞ
X ðkÞ ¼ x1 ; x2 ; . . . . . . . . . ; xNC ðA:4Þ Some of the variables may cross the lower or upper bounds in
ðkÞ ðkÞ ðkÞ updated population X modified
i after teacher phase then it is set to
ðkÞ x1j þx2j þþxNPj
where xj ¼ NP
for j = 1,2,. . ., NC subjects its limiting values.
874 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

Table E.1
Line data.

Line no. From bus no. To bus no. Line parameter Line flow limit (p.u.)
ri (p.u.) xi (p.u.) Charging susceptance (p.u.) Tap ratio
1 1 2 0.0035 0.0411 0.6987 – 5.054
2 1 39 0.0010 0.0250 0.7500 – 5.8225
3 2 3 0.0013 0.0151 0.2572 – 9.5191
4 2 25 0.0070 0.0086 0.1460 – 1.2746
5 3 4 0.0013 0.0213 0.2214 – 3.8055
6 3 18 0.0011 0.0133 0.2138 – 1.6741
7 4 5 0.0008 0.0128 0.1342 – 1.6094
8 4 14 0.0008 0.0129 0.1382 – 3.2224
9 5 6 0.0002 0.0026 0.0434 – 2.662
10 5 8 0.0008 0.0112 0.1476 – 2.0368
11 6 7 0.0006 0.0092 0.1130 – 3.1735
12 6 11 0.0007 0.0082 0.1389 – 5.3507
13 7 8 0.0004 0.0046 0.0780 – 1.2782
14 8 9 0.0023 0.0363 0.3804 – 8.0844
15 9 39 0.0010 0.0250 1.2000 – 8.2768
16 10 11 0.0004 0.0043 0.0729 – 5.4104
17 10 13 0.0004 0.0043 0.0729 – 3.7891
18 13 14 0.0009 0.0101 0.1723 – 3.7369
19 14 15 0.0018 0.0217 0.3660 – 1.2315
20 15 16 0.0009 0.0094 0.1710 – 4.9478
21 16 17 0.0007 0.0089 0.1342 – 1.6503
22 16 19 0.0016 0.0195 0.3040 – 6.233
23 16 21 0.0008 0.0135 0.2548 – 4.2317
24 16 24 0.0003 0.0059 0.0680 – 1.4786
25 17 18 0.0007 0.0082 0.1319 – 1.4598
26 17 27 0.0013 0.0173 0.3216 – 1.561
27 21 22 0.0008 0.0140 0.2565 – 8.3709
28 22 23 0.0006 0.0096 0.1846 – 1.1836
29 23 24 0.0022 0.0350 0.3610 – 4.8179
30 25 26 0.0032 0.0323 0.5130 – 3.543
31 26 27 0.0014 0.0147 0.2396 – 4.9731
32 26 28 0.0043 0.0474 0.7802 – 2.2766
33 26 29 0.0057 0.0625 1.0290 – 3.2294
34 28 29 0.0014 0.0151 0.2490 – 5.1174
35 12 11 0.0016 0.0435 0.0000 1.006 0.63118
36 12 13 0.0016 0.0435 0.0000 1.006 0.76421
37 6 31 0.0000 0.0250 0.0000 1.070 1.5159
38 10 32 0.0000 0.0200 0.0000 1.070 9.3063
39 19 33 0.0007 0.0142 0.0000 1.070 8.8705
40 20 34 0.0009 0.0180 0.0000 1.009 7.386
41 22 35 0.0000 0.0143 0.0000 1.025 9.2532
42 23 36 0.0005 0.0272 0.0000 1.000 7.8687
43 25 37 0.0006 0.0232 0.0000 1.025 7.7447
44 2 30 0.0000 0.0181 0.0000 1.025 3.5378
45 29 38 0.0008 0.0156 0.0000 1.025 12.076
46 19 20 0.0007 0.0138 0.0000 1.060 2.6197

At the end of teacher phase if X modified

i gives minimum value of Otherwise
objective function then retain the updated value otherwise previ- ðkÞ
ðkÞ
ous population X i is considered as input population for student X new
i ¼ X i þ rand  ½X j  X i  ðA:8Þ
phase. Thus better population as per above criteria is stored in In case of limit violation the variables are set to their limiting
ðkÞ
X i and this updated population becomes the input for learner values.
phase. If X new
ðkÞ ðkÞ
is better than X i ; X i is replaced by X new
ðkÞ
, otherwise X i
i i
is retained in new population at the end of learner phase. This pro-
Appendix A.3. Learner phase cedure is repeated from relation (A.3) till a maximum number of
iterations are executed say kmax = 500.
As learner also learns the subjects with mutual discussions
and interacting with other learners in the class based on Appendix B. Overview of modified version of bare bones
the other students knowledge about the subjects. So there may particle swarm optimization (BBExp)
be possibility for a learner to improve upon the results in
subjects with this interaction. This interaction is assumed as The Bare bones particle swarm optimization (BBPSO) is a popu-
random. lation based optimization technique developed by Kennedy [26]
Therefore in this phase randomly two learners Xi and Xj, i – j are with Gaussian distribution to update a particle using mean and
ðkÞ
selected out of updated population X i as obtained after teacher standard deviation based on best position of particle and global
phase learning. Thus new set of learners at the end of learner phase best position. At each iteration ‘k’ the ith particle is represented
is obtained as given below for i = 1,. . ., NP. ðkÞ
by a vector X i : Thus BBPSO maintain a set of positions as follows:
ðkÞ
X new
i ¼ X i þ rand  ½X i  X j ; ifJðX i Þ < JðX j Þ ðA:7Þ ðkÞ ðkÞ ðkÞ
SðkÞ ¼ ½X 1 ; X 2 ; . . . ; X NP  ðB:1Þ
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 875

Table E.2
Load data.

Bus no. Bus voltage limit Vi (p.u.) Base case load condition (without load shed) Predicted load condition (without load shed)

Vi Vi PD (p.u.) QD (p.u.) PD (p.u.) QD (p.u.)

1 0.97 1.1 0 0 0 0
2 0.97 1.1 0 0 0 0
3 0.97 1.1 3.542 0.0264 3.703 0.0276
4 0.97 1.1 5.5 2.024 5.75 2.116
5 0.97 1.1 0 0 0 0
6 0.97 1.1 0 0 0 0
7 0.97 1.1 2.5718 0.924 2.6887 0.966
8 0.97 1.1 5.742 1.936 6.003 2.024
9 0.97 1.1 0 0 0 0
10 0.97 1.1 0 0 0 0
11 0.97 1.1 0 0 0 0
12 0.97 1.1 0.0825 0.968 0.08625 1.012
13 0.97 1.1 0 0 0 0
14 0.97 1.1 0 0 0 0
15 0.97 1.1 3.52 1.683 3.68 1.7595
16 0.97 1.1 3.619 0.3553 3.7835 0.37145
17 0.97 1.1 0 0 0 0
18 0.97 1.1 1.738 0.33 1.817 0.345
19 0.97 1.1 0 0 0 0
20 0.97 1.1 6.908 1.133 7.222 1.1845
21 0.97 1.1 3.014 1.265 3.151 1.3225
22 0.97 1.1 0 0 0 0
23 0.97 1.1 2.7225 0.9306 2.8463 0.9729
24 0.97 1.1 3.3946 1.012 3.5489 1.058
25 0.97 1.1 2.464 0.5192 2.576 0.5428
26 0.97 1.1 1.529 0.187 1.5985 0.1955
27 0.97 1.1 3.091 0.8305 3.2315 0.86825
28 0.97 1.1 2.266 0.3036 2.369 0.3174
29 0.97 1.1 3.1185 0.2959 3.2602 0.30935
30 – – 0 0 0 0
31 – – 0.1012 0.0506 0.1058 0.0529
32 – – 0 0 0 0
33 – – 0 0 0 0
34 – – 0 0 0 0
35 – – 0 0 0 0
36 – – 0 0 0 0
37 – – 0 0 0 0
38 – – 0 0 0 0
39 – – 12.144 2.75 12.696 2.875

Table E.3
Generator data.

Generator no. Bus no. Voltage magnitude (p.u) Real generation schedule (p.u.) Reactive generation (p.u.)
Maximum Minimum
1 39 1.0300 10 16.0 2.00
2 31 0.9820 0 1.5 0.20
3 32 0.9831 6.50 8.0 1.5
4 33 0.9972 6.32 8.0 1.5
5 34 1.0123 5.08 6.0 1.5
6 35 1.0493 6.50 7.0 1.5
7 36 1.0635 5.60 6.5 1.5
8 37 1.0278 5.40 6.0 1.5
9 38 1.0265 8.30 10.0 2.0
10 30 1.0475 2.50 3.0 0.3

ðkÞ
Each particle has its best position Pbesti referred to as personal A computational procedure for optimization using BBExp is
best found by the particle so far, and the very best position explained in following steps:
ðkÞ
amongst all particle referred to global best Gbest .
Step-(a): A initial population of size ‘NP’ is generated as follows:
ðkÞ ðkÞ ðkÞ ðkÞ
PðkÞ ¼ ½Pbest1 ; P best2 ; . . . ; Pbesti ; . . . ; PbestNP  ðB:2Þ
ð0Þ ð0Þ ð0Þ
ðkÞ ðkÞ
Sð0Þ ¼ ½X 1 ; X 2 ; . . . . . . X NP  ðB:4Þ
Further one denotes the P besti of the best particle in S as Gbest .
ð0Þ ð0Þ ð0Þ ð0Þ T
ðkÞ ðkÞ Xi ¼ ½xi1 ; xi2 ; :::::xiNC  ðB:5Þ
Gbest ¼ arg min JðP besti Þ i ¼ 1; 2; . . . ; NP ðB:3Þ
ð0Þ
xij
i.e. jth parameter of Xi vector is obtained from uniform distri-
J() is the objective function. bution as follows
876 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877

ð0Þ
xij ¼ xjmin þ ðxjmax  xjmin Þui ðB:6Þ g l ðX p Þ if g l ðX p Þ > 0
Gl ðX p Þ ¼ ðC2Þ
¼ 0 otherwise
xj-min and xj-max are lower and upper bounds on variable xj. ui is a
random digit in the range [0, 1]. where g l ðXÞ 6 0; l ¼ 1; . . . ; L are inequality constraints.
Step-(b): Set iteration count k = 1. Calculate violation qp and qq
ðkÞ ðkÞ
Step-(c): Obtain P besti and Gbest based on objective function. !,
X
L
Step-(d): Obtain modified particle as follows qp ¼ Gl ðX p Þ L ðC3Þ
l¼1
ðkÞ ðkÞ
Pbestij þ Gbestj !,
lðkÞ
ij ¼ for all i and j ðB:7Þ X
L
2 qq ¼ Gl ðX q Þ L ðC4Þ
l¼1
rðkÞ ðkÞ ðkÞ
ij ¼ jP bestij  Gbestj j for all i and j ðB:8Þ
If qp < qq
ðkÞ ðkÞ
where lij and rij are the mean and absolute difference of Pbest Then Xp is better than Xq, otherwise Xq is better than Xp.
and Gbest in jth decision variable of ith particle at kth iteration
respectively. Updated particle is given as Appendix D. Appendix
ðkþ1Þ ðkÞ ðkÞ
xij ¼ Nðl ij ; r ij Þ ðB:9Þ
IEEE-30 bus system data (100 MVA base) [28].
ðkÞ ðkÞ See Tables D.1–D.4.
Nðl r ij ; ij Þ represents a normally distributed random variate of
ðkÞ ðkÞ
mean l ij and standard deviation as ij . This has been termedr as
Appendix E. Appendix
BBPSO.
ðkÞ ðkÞ ðkÞ
For P bestij ¼ Gbestj results r ij ¼ 0 and hence no updation in
ðkþ1Þ
IEEE-39 bus system data (100 MVA base) [29].
particle xij : Thus it may cause premature convergence. See Tables E.1–E.3.
An alternative version of BBPSO also, referred to as BBExp [27] is
used in this paper. Where the jth dimension of the ith particle is References
updated as follows:
8 9 [1] Stott B, Alsac O, Monticelli AJ. Security analysis and optimization. Proc IEEE
< NðlðkÞ ; rðkÞ Þ if U½0; 1 < 0:5 = 1987;75(12):1623–44.
ðkþ1Þ ij ij
xij ¼ ðB:10Þ [2] Hajdu LP, Peschon J, Tinney WF, Piercy DS. Optimal load shedding policy for
: PðkÞ ; power systems. IEEE Trans PAS 1968;87(3):784–95.
bestij
[3] Palaniswamy KA, Misra KB, Sharma J. Optimum load shedding taking into
account voltage and frequency characteristics of loads. IEEE Trans PAS
where U[0, 1] is a uniformly distributed random digit between 1985;104(6):1342–8.
range 0 and 1. [4] Medicherla TKP, Billinton R, Sachdev MS. Generation rescheduling and
This means there is a 50% chance that the jth dimension of the overloads analysis. IEEE Trans PAS 1979;98(6):1876–84.
[5] Shah S, Shahidehpour SM. A heuristic approach to load shedding scheme. IEEE
ith particle changes to the corresponding Pbest of the particle.
Trans Power Syst 1989;4(4):1421–9.
Therefore, particles updated by BBExp algorithm are allowed [6] Berg GJ, Sharaf TA. System loadability and load shedding. Electr Power Syst Res
to take some of the variables in their Pbests. Thus, BBExp is said 1994;28(3):217–25.
to be biased toward exploiting the Pbest positions. [7] Tuan TQ, Fandino J, Hadjsaid N, Sabonnadiere JC, Vu H. Emerging load shedding
to avoid risk of voltage instability using indicators. IEEE Trans Power Syst
1994;9(1):341–51.
Step-(e): The resulting positions of each individual may not sat- [8] Bijwe PR, Tare RS, Kelapure SM. Anticipatory load shedding scheme for
isfy constraints on decision variables, in such situations deci- loadability enhancements. IEEE Proc GTD 1999;146(3):483–90.
[9] Jung J, Liu CC, Tanimoto SL, Vittal V. Adaption in load shedding under
sion variables are set to their limiting values. vulnerable operating conditions. IEEE Trans Power Syst
ðkþ1Þ ðkþ1Þ
Step-(f): P besti and Gbest are obtained as follows 2002;17(4):1199–2005.
[10] Echavarren FM, Lobato E, Rouco L. A corrective load shedding scheme to
ðkþ1Þ ðkþ1Þ ðkþ1Þ ðkÞ mitigate voltage collapse. Int J Electr Power Energy Syst 2006;28(1):58–64.
Pbesti ¼ X i if f ðX i Þ < f ðPbesti Þ [11] Chattopadhyay D, Chakrabarti BB. A preventive/corrective model for voltage
ðkÞ
ðB:11Þ stability incorporating dynamic load shedding. Int J Electr Power Energy Syst
¼ Pbesti Otherwise 2003;25(5):363–76.
[12] Pande VS, Arya LD. Steady state load shedding for line overload alleviation
where f() is the function to be optimized. Further global best using Hopefield model based optimization. J Inst Eng (India) 2005;86:142–8.
ðkþ1Þ pt-EL, September.
particle of swarm i.e. Gbest is set as
[13] Amraee T, Ranjbar AM, Mozafari B, Sadati’ N. An enhanced under-voltage load-
ðkþ1Þ ðkþ1Þ shedding scheme to provide voltage stability. Electr Power Syst Res
Gbest ¼ arg  min f ðPbesti Þ i ¼ 1 . . . NP ðB:12Þ 2007;77(8):1038–46.
[14] Girgis AA, Mathure S. Application of active power sensitivity to frequency and
Step-(g): Objective function is evaluated for updated vector and voltage variations on load shedding. Electr Power Syst Res 2010;80(3):306–10.
increase k = k + 1, and repeat from Step-(c) till convergence for a [15] Fu Xu, Wang Xifan. Determination of load shedding to provide voltage
maximum number of generation (kmax). The procedure may stability. Int J Electr Power Energy Syst 2011;33(3):515–21.
[16] Amraee T, Ranjbar AM, Feuillet R. Adaptive under-voltage load shedding
even terminated if there is no improvement in the objective scheme using model predictive control. Electr Power Syst Res
function for a specified number of generations is observed. 2011;81(7):1507–13.
[17] Arya LD, Singh P, Titare LS. Differential evolution applied for anticipatory load
Appendix C. Handling of inequality constraints
shedding with voltage stability considerations. Int J Electr Power Energy Syst
2012;42(1):644–52.
(i) If two vectors satisfy inequality constraints than Xp will be [18] Gu W, Liu W, Shen C, Wu Z. Multi-stage underfrequency load shedding for
islanded microgrid with equivalent inertia constant analysis. Int J Electr Power
better than Xq if
Energy Syst 2013;46:36–9.
[19] Sigrist L, Egido I, Rouco L. A method for the design of UFLS schemes of small
JðX p Þ < JðX q Þ ðC1Þ isolated power systems. IEEE Trans Power Syst 2012;27(2):951–8.
[20] Karimi M, Mohamad H, Mokhlis H, Bakar AHA. Under-frequency load shedding
where J() is optimization function. scheme for islanded distribution network connected with mini hydro. Int J
(ii) Otherwise evaluate Electr Power Energy Syst 2012;42(1):127–38.
 L.D. Arya, A. Koshti / Electrical Power and Energy Systems 63 (2014) 862–877 877

[21] Hooshmand R, Moazzami M. Optimal design of adaptive under frequency load [25] Rao RV, Savsani VJ, Vakharia DP. Teaching–learning-based optimization: an
shedding using artificial neural networks in isolated power system. Int J Electr optimization method for continuous non-linear large scale problems. Int J Inf
Power Energy Syst 2012;42(1):220–8. Sci 2012;183(1):1–15.
[22] Goh HH, Kok BC, Yeo HT, Lee SW, Mohd Zin AA. Combination of TOPSIS and [26] Kennedy J. Bare bones particle swarms. In: Proceeding of the IEEE swarm
AHP in load shedding scheme for large pulp mill electrical system. Int J Electr intelligence symposium; 2003. p. 80–7.
Power Energy Syst 2013;47:198–204. [27] Zhang H, Kennedy DD, Rangaiah GP, Bonilla-Petriciolet A. Novel bare-bones
[23] Arief Ardiaty, Dong Zhao Yang, Nappu Muhammad Bachtiar, Gallagher Marcus. particle swarm optimization and its performance for modeling vapor–liquid
Under voltage load shedding in power systems with wind turbine-driven equilibrium data. Fluid Phase Equilib 2011;301(1):33–45.
doubly fed induction generators. Electr Power Syst Res 2013;96:91–100. [28] Pai MA. Computer techniques in power system analysis. second ed. TMH
[24] Sun Wei-Qing, Zhang Yan, Wang Cheng-Min, Song Ping. Publication; 2010.
Flexible load shedding strategy considering real-time dynamic thermal line [29] Padiyar KR. Power system dynamics stability and control. second ed. BS
rating. IET GTD 2013;7(2):130–7. Publication; 2008.