Energy Conversion and Management 75 (2013) 130–140

Contents lists available at SciVerse ScienceDirect

Energy Conversion and Management

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

Application of computational intelligence techniques for load sheddingin power systems: A review

0196-8904/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.enconman.2013.06.010

⇑ Corresponding author at: Department of Electrical Engineering, Faculty ofEngineering, University of Malaya, 50603 Kuala Lumpur, Malaysia. Tel.: +60 379675238; fax: +60 03 79675316.

E-mail address: javedahmedleghari@gmail.com (J.A. Laghari).

J.A. Laghari a,d,⇑, H. Mokhlis a,b, A.H.A. Bakar b, Hasmaini Mohamad c

a Department of Electrical Engineering, Faculty of Engineering, University of Malaya, 50603 Kuala Lumpur, Malaysiab University of Malaya Power Energy Dedicated Advanced Centre (UMPEDAC), Level 4, Wisma R&D UM, Jalan Pantai Baharu, University of Malaya, 59990 Kuala Lumpur, Malaysiac Faculty of Electrical Engineering, University of Technology MARA (UiTM), 40450 Shah Alam, Selangor, Malaysiad Department of Electrical Engineering, Quaid-e-Awam University of Engineering Science & Technology, Nawabshah, 67480 Sindh, Pakistan

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

Article history:Received 4 April 2013Accepted 7 June 2013

Keywords:Load sheddingArtificial neural networkFuzzy logic controlAdaptive neuro-fuzzy inference systemGenetic algorithmParticle swarm optimization

Recent blackouts around the world question the reliability of conventional and adaptive load sheddingtechniques in avoiding such power outages. To address this issue, reliable techniques are required to pro-vide fast and accurate load shedding to prevent collapse in the power system. Computational intelligencetechniques, due to their robustness and flexibility in dealing with complex non-linear systems, could bean option in addressing this problem. Computational intelligence includes techniques like artificial neuralnetworks, genetic algorithms, fuzzy logic control, adaptive neuro-fuzzy inference system, and particleswarm optimization. Research in these techniques is being undertaken in order to discover means formore efficient and reliable load shedding. This paper provides an overview of these techniques as appliedto load shedding in a power system. This paper also compares the advantages of computational intelli-gence techniques over conventional load shedding techniques. Finally, this paper discusses the limitationof computational intelligence techniques, which restricts their usage in load shedding in real time.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Over the last decades, the world has witnessed many severepower system blackouts in various continents. These blackoutshave affected millions of people, resulting in huge economic lossand social impact over the societies. The social impact of a powerblackout, especially in urban areas, is severe – health care facilitiesin hospitals are affected, traffic control problems lead to accidents,and the Internet and other communications systems break down.The detailed consequences of power blackouts are shown inFig. 1 [1].

A blackout in a power system refers to the unavailability ofelectric power in an area for a short or long duration. These powerblackouts can occur due to natural reasons as well as technical rea-sons. Natural reasons include animal contact with a live conductor,a vehicular accident resulting in damaged transmission poles, andtrees falling on transmission lines due to stormy weather. Techni-cal reasons include faults, damaged transmission or distributionlines, stability issues, overloaded transmission lines, cascadingevents, faulty equipment, and human error. The estimated unsup-plied energy is another important factor leading to blackouts. Some

of these reasons, like faults, are initiating contingencies; others arethe subsequent consequences of those events which may result ininstability and cascading, leading to a blackout. This paper focuseson those power blackouts which have occurred due to technicalfailure. The top ten most severe power blackouts that have oc-curred in the last two decades, affecting millions of people, areshown in Table 1.

Table 1 shows that within this period, Brazil and India suffereda large-scale blackout at least twice, whereas the other countries/regions like Egypt and Europe were hit by a massive blackout onlyonce, in 1990 [2], and in 2006 [6,7], respectively. During theseblackouts, 50 million people in Egypt were affected for 6 h, while15 million people in Europe were affected for 2 h. The blackoutin Egypt was characterized by a very rapid voltage collapse tonearly 20 V, followed by sudden total voltage collapse. The severeblackout that occurred in India in 2001 was due to a failure of sub-stations. This blackout affected 226 million people for 12 h and re-sulted in an overall economic loss of 110 million USD. One of themost significant blackouts occurred in the United States and Can-ada on 14th August 2003. This blackout affected around 50 millionpeople in eight US states and two Canadian provinces. Estimatesshow that this blackout interrupted around 63 GW of load, andmore than 400 transmission lines and 531 generating units at261 power plants tripped [4,5,9]. It lasted for 96 h (4 days) in var-ious parts of the eastern United States [10] as shown in Table 1,resulting in an economic loss of approximately 4–6 billion USD

Consequences of Blackouts

Morbidity Increase

Lost of Trust in Banks

Road Traffic Restrictions

Telephone Network Collapse

Internet Breakdown

Mass Litigation

Mortality Increase

Health Care and Treatment Insufficient

Payment Transaction Restrictions

Restrictions of Medical Facilities

Stock Market Bust

Business Interruptions in Manufacturing Process Rail/Traffic Restrictions

Fig. 1. Consequences of power blackouts.

Table 1Most severe power blackout in last two decades around the world.

Country Date Affected people Duration Causes of blackout

Egypt 24th April 1990 50 million [2] 6 h Voltage collapseBrazil 11th March 1999 97 million [3] 5 h Lightning strike causing 440 kV circuits to tripIndia 2nd January 2001 226 million [3] 12 h Transmission line faultCanada and Northeast United states 14th August 2003 55 million [4,5] 96 h (4 days) Lack of maintenance, human error and equipment failureItaly 28th September 2003 56 million [5] 18 h Tripping of power linesIndonesia 18th August 2005 100 million [3] 7 h Transmission line failureEurope 4th November 2006 15 million [6,7] 2 h OverloadingBrazil and Paraguay 10th November 2009 87 million [3]. 7 h Short circuit on three transformers on high voltage transmission lineBrazil 4th February 2011 53 million [3] 16 h Flaw in transmission lineIndia 31st July 2012 670 million [8] 15 h Voltage collapse due to overloading of transmission line

J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140 131

[11]. Indonesia suffered a severe blackout in 2005 that affected100 million people for 7 h [3]. The world’s largest blackout hap-pened recently on 31st July 2012 in India following a voltage col-lapse due to the overloading of transmission lines. It affectedaround 670 million people, hundreds of trains, and hundreds ofthousands of households in 22 Indian states [8].

Apart from these severe blackouts, every country suffers fromsmall power outages many times in a year. Fig. 2 shows the num-ber of power outages that occurred in different parts of the worldin 2009; Fig. 3 shows the duration of these power outages. Thisstudy was conducted by the Sustainable Development Networkof the World Bank Group [12].

Figs. 2 and 3 shows that South Asia had up to 1200 power out-ages, but these had the shortest duration compared to those inother parts of the world. Latin America and the Caribbean experi-enced the fewest power outages, but their duration was the longest

Fig. 2. Number of power outages

compared to those in other places. According to an annual reportissued by the Eaton Blackout Tracker in 2011, there were 3071power outages in different states of the US that year, affecting41.8 million people. The top ten states in the US with the mostnumber of reported power outages in 2011 are shown in Fig. 4.

It can be observed that many developed states in the US, such asWashington, New Jersey, Michigan, Texas, New York, and Califor-nia, also had a significant number of power outages in 2011 [1].

The most common factor contributing to power blackouts is thevoltage instability issue arising from the overloading of the trans-mission system [2], which may result in a cascading or islandingevent leading to a blackout, as in the Egypt and India blackouts.During such conditions, accurate load shedding is crucial to pre-vent total system collapse. However, improper load shedding hasled to a high number of power blackouts due to surplus or insuffi-cient load shed; this has questioned the ability and reliability of

in different parts of world.

Fig. 3. Duration of power outages occurred in different parts of world.

Fig. 4. Top ten USA states with most reported power outages in 2011.

132 J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140

existing conventional load shedding techniques. Hence, alternativetechniques are required to enhance the reliability of today’s mod-ern, complex, and large power systems.

Computational intelligence techniques have attracted theresearchers’ attention due to their robustness and ability to dealwith complex systems easily. Various researchers have proposedcomputational intelligence techniques for addressing load shed-ding problems. This paper presents a review on the ability of thesetechniques to protect a power system from blackouts.

This paper is organized as follows: Section 2–3 discusses fre-quency control in the normal operation of a power system, loadshedding techniques and their respective types. Sections 4–5provide a review of applications for computational intelligencetechniques (CIT) in load shedding and discussion. Finally, the con-clusion is presented.

2. Frequency control in a power system

The stable operation of a power system requires frequency andvoltage to be constant. In practice, the frequency in a power systemis never in a balanced state since the load demand varies continu-ously. In an electric power system, the power generated must bekept in constant equilibrium with the power consumed; otherwise,a power deficiency will occur. The system frequency decreases ifload exceeds generation and increases when power generation isgreater than load demand [13]. Power system frequency is directlyproportional to generator speed [14]. Hence, one way of controllingthe frequency is by regulating the generator speed. The generatoris commonly equipped with a governor for sensing and monitoringthe speed continuously. Controlling the frequency in an isolatedpower system having a single generator is much easier than thatin an interconnected power system. When the load suddenly in-creases in an isolated power system having a single generator,

the extra energy demand is initially supplied by the rotational iner-tia of the generator. The rotational speed of the generator then de-creases, which results in a proportional decrease in the systemfrequency. The governor opens the turbine gate in order to increasethe turbine speed. The increase in turbine speed will increase thesystem frequency; thus, frequency can be recovered within accept-able range.

For frequency control in an interconnected power system,power utilities employ a controlling mechanism to recover the fre-quency during transient faults or severe load variations. Fig. 5shows the various control actions required to recover frequencyin order to avoid a power blackout [15].

It can be observed from Fig. 5 that in case of power deviation,primary control action is activated to re-establish the balance be-tween load demand and generation. The set point for this controlaction is 50 Hz. Any deviation from this set point will cause the pri-mary controllers of all the generators to respond within a few sec-onds (<15 s). The controller changes the power output of thegenerators until a balance between power output and load demandis re-established. After 15 s, the remaining frequency and powerdeviation will be controlled by activating the secondary control.The function of the secondary control is to restore power and fre-quency to their nominal values. In order to maintain this balance,generation capacity for use as a secondary reserve must be avail-able to cover any outages or any disturbances affecting production,consumption, and transmission. The secondary control operates forperiods lasting several minutes. In the end, the remainingfrequency and power deviation is supplied by activating the ter-tiary control actions. An automatic or manual change in the work-ing points of the generators or the participating loads is required inorder to guarantee the provision of an adequate secondary controlreserve at the right time. The power which can be connected auto-matically or manually under tertiary control is known as the

System Frequency

Secondary Control

Tertiary Control

Primary Control

Time Control

Limit Deviation

Restore Normal

Activate

Restore Mean

Free Reserve

Free Reserve after outage

Free ReserveCorrect

Activate on Long Term

Activate if Responsible

Take over if Responsible

Take over

Fig. 5. Frequency control in power system.

J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140 133

tertiary control reserve. If the system’s mean frequency in the syn-chronous zone deviates from the nominal frequency of 50 Hz,resulting in a discrepancy between synchronous time and univer-sal coordinated time. This time offset serves as a performance indi-cator for the primary, secondary, and tertiary controls and mustnot exceed 30 s. Nowadays, power system networks mostly oper-ate closer to their stability limits as a consequence of the deregu-lated electricity market and growth in energy consumption.Under such operating conditions, a severe disturbance such as aloss of generating units or faults along transmission lines may leadto cascading events. Thus, the risk of collapse of the overall powersystem is increased. In order to address such issues in which fre-quency declines very fast and goes below the specific threshold va-lue, an under frequency load shedding technique is applied torecover the system frequency and avoid a complete power black-out [16].

3. Load shedding techniques and their types

Load shedding techniques are commonly divided into threemain categories – conventional, adaptive, and computational intel-ligence-based. Fig. 6 illustrates the different types that fall undereach category:

Load Shedding Techniques

Conventional Load Shedding Techniques

Under Frequency Load Shedding Techniques

Under Voltage Load Shedding Techniques

Fuzzy Cont

Artificial Neural Networks (ANN)

Adaptive Load Shedding Techniques

Fig. 6. Types of load sh

3.1. Conventional load shedding techniques

Conventional load shedding techniques are of two types:

3.1.1. Under frequency load shedding (UFLS) techniquesUnder frequency load shedding is applied in the case of a severe

fault, faster decrease in frequency due to the loss of generators.According to Institute of Electrical and Electronics Engineers (IEEE)standards, ‘‘under frequency load shedding must be performedquickly to arrest power system frequency decline by decreasingpower system load to match available generating capacity’’ [17].For this purpose, certain frequency threshold values are set to startthe under frequency load shedding. The minimum acceptable fre-quency is dependent on the system equipment, such as the gener-ator type, its auxiliary device, and the turbine [18]. The UFLS relayis initialized to shed a fixed amount of load in predefined stepswhen frequency falls below a certain predefined threshold in orderto prevent a blackout [16]. The European Network of TransmissionSystem Operators for Electricity (ENTSOE) has recommended thefollowing steps for under frequency load shedding [15]:

(1) The first stage of automatic load shedding should be initi-ated at 49 Hz.

(2) At 49 Hz, at least 5% of total consumption should be shed.

Computational Intelligent Load Shedding Techniques

Logic rol

Adaptive Neuro Fuzzy Inference System (ANFIS)

Particle Swarm Optimization

Genetic Algorithm

edding techniques.

Measure Frequency / Voltage

End

Frequency = 50 Hz/

Voltage within range

No

Activate Under frequency Relay

/Under voltage Relay

Shed load

Start

Yes

No

Yes

Frequency < f/

Voltage < V

min

min

Fig. 7. Flow chart of conventional load shedding techniques.

134 J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140

(3) A stepwise 50% of the nominal load should be disconnectedby using under frequency relays in the frequency range of49.0–48.0 Hz.

(4) In each step of load shedding, a disconnection of no morethan 10% of the load is advised.

(5) The maximum disconnection delay should be 350 msincluding breakers’ operating time.

3.1.2. Recommendations for power plantsThe following recommendations are necessary for the safe oper-

ation of power plants [15]:

(1) At 49.8 Hz, ‘quick-start’ plants should be connected to thegrid.

(2) For a power system operating within 50 Hz (60 Hz) fre-quency, the minimum allowable operating frequency usu-ally specified by the manufacturer according to the turbinetype is 47.5 Hz (57.5 Hz) [19–21]. This is necessary for theprotection of the generator and its auxiliary equipmentbecause power plant auxiliary services begin to malfunctionat a frequency of 47.5 Hz; the situation becomes critical atabout 44–46 Hz. Furthermore, generator operation at47.5 Hz or below could damage the turbine blades andreduce its lifespan [22]. Hence, load shedding in a power sys-tem helps to prevent the loss of generators, equipment dam-age, and blackouts.

3.1.3. Under voltage load shedding (UVLS) techniquesUVLS techniques are implemented to protect the power system

from voltage collapse. A look at major power blackouts that haveoccurred around the world show that most were caused by voltageinstability problems [2]. Voltage instability generally occurs due toeither forced outage of the generator or the line, or overloading.When this happens, the reactive power demand in transmissionlines varies severely and may cause a blackout if not recoveredquickly. The UVLS technique is applied by power utilities to pre-vent voltage instability and restore voltage to its nominal value[23].

The flow chart for under frequency and under voltage loadshedding techniques is shown in Fig. 7 [20,21].

3.1.4. Limitation of conventional load shedding techniquesConventional load shedding techniques are limited by their

inability to provide optimum load shedding. They simply followa preset rule in which a fixed amount of load is shed when fre-quency deviates from the nominal value. The main disadvantageof this method is that it does not estimate the actual amount ofthe power imbalance. The result is either over-shedding, which af-fects power quality, or under-shedding, which leads to tripping ofelectricity service [16].

3.2. Adaptive load shedding techniques

Adaptive load shedding techniques employ a power swingequation to shed the required amount of load. The power imbal-ance within the system can be obtained by using this equation[24]:

DP ¼ 2Hf� @f@t

ð1Þ

where DP is the power imbalance, H is inertia constant of generator,f is nominal frequency (Hz), and df/dt is the rate of change of fre-quency (Hz/s).

This equation can be applied to an isolated power system hav-ing only a single generator as well as to an interconnected power

system. Whenever, the system suffers a disturbance (fault or islan-ding), there is variation in frequency as well as rate of change offrequency (ROCOF). By putting these values in Eq. (1), powerimbalance can be estimated. After estimating the power imbal-ance, the required amount of load is shed in order to stabilizethe frequency. The most common example of adaptive load shed-ding is the ROCOF relay. The performance of adaptive load shed-ding techniques can be improved by using both frequency andvoltage deviations. Such a technique is proposed in [16], whichshows that the proposed technique can enhance the reliability interms of frequency stability and voltage stability, displaying goodtransient behavior when encountering severe disturbances. Theflow chart for adaptive load shedding techniques is shown inFig. 8 [20,21].

Adaptive load shedding techniques enhance the reliability ofconventional load shedding. However, these techniques also sufferfrom un-optimum load shedding due to variations in df/dt behav-ior. The df/dt value has been found to depend upon the operating

Measure Frequency

End

No

Measure Power imbalance by using power swing

equation

Shed estimated load

Start

Yes

No

Yes

minFrequency < f

Frequency = 50 Hz

Fig. 8. Flow chart of Adaptive load shedding techniques.

Table 2Comparison features of conventional and computational intelligence techniques.

No Feature Conventional techniques Computationalintelligence techniques

1 Optimumloadshedding

Do not provide optimum loadshedding

Have the ability toprovide optimum loadshedding

2 Complexpowersystem

Cannot deal efficiently withmodern and complex powersystems

Can deal efficiently withmodern and complexpower systems

J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140 135

capacities of a power system (base load capacity, peak load capac-ity). The values of df/dt are different for similar amounts of loadvariation at base and peak capacity. This variation in df/dt behaviorresults in the un-optimum estimation of the power imbalance andaffects the performance of adaptive load shedding techniques.

3.3. Computational intelligence based load shedding techniques

The term ‘computational intelligence techniques’ generally re-fers to a set of techniques that are applied to mimic human intel-ligence. These techniques include artificial neural networks (ANN),adaptive neuro-fuzzy inference system (ANFIS), fuzzy logic control(FLC), genetic algorithms (GA), and particle swarm optimizations(PSO). These techniques can easily solve those nonlinear, multi-objective problems in power systems that cannot be solved bythe conventional methods with the desired speed and accuracy[25,26].

Due to the modern power system’s complex structure and largesize, traditional load shedding techniques may not work efficientlyin a contingency. A conventional UFLS technique may shed unnec-essary and extra loads because it disconnects the load in fixedsteps without actually measuring the power imbalance [27]. Inad-equate load shedding causes the frequency to vary significantly,which may then lead to a blackout, whereas disconnecting more

load than required will result in excessive power outage [28–30].Furthermore, the optimal load shedding technique is a nonlinearoptimization problem dealing with multiple constraints. Conven-tional optimization techniques have been proven to be inadequatewhen dealing with complex non-linear problems [31]. Thus, anefficient load shedding technique is crucial to shed optimum loadand maintain power system stability.

Before actual implementation on a real power system, thesecomputational intelligence techniques go through a series of sim-ulations to determine the optimum load shedding for various con-tingencies such as, faults, line tripping, voltage instability issues,splitting of the power system into different islands, and frequencystability issues. After the successful training and testing or optimi-zation of the techniques for these scenarios, they are applied inreal-time conditions. If the power system suffers from any of theabove problems, these techniques can provide optimum load shed-ding for that case, as the optimum solution for that case has al-ready been determined.

The advantages of computational intelligence techniques overconventional techniques are summarized in Table 2.

4. Applications of computational intelligence techniques (CIT)for load shedding in power systems

Since the late 1980s, the attraction of using computationalintelligence techniques in power systems has increased. CIT havebeen widely applied in power system applications. The implemen-tation of each CIT technique for load shedding in a power system,along with its advantages and disadvantages, is discussed below:

4.1. Artificial neural network (ANN) application in load shedding

An ANN is a mathematical model based on human neural sys-tems. It has been widely used in power system problems. ANNapplication in power systems include voltage stability [32], systemsecurity [33–35], dynamic stability [36], steady state stability [37–39], transient stability [40,41], load forecasting [42], harmonicmonitoring [43], and transmission lines protection [44,45].

Various researchers have applied ANN to load shedding in apower system. Hsu et al. [28] have proposed an ANN-based loadshedding technique for cogeneration systems. The ANN trainingprocess was organized by considering three inputs – total genera-tion, total load demand, and frequency decay rate – and one output– minimum amount of load shedding. The technique was verifiedand compared with the conventional technique. The results showthat the proposed technique performed load shedding morequickly as compared with the conventional technique [28]. Otherapplications of ANN to provide quick and optimum load sheddingin an isolated power system are presented in [27,46]. The tech-nique was verified on a 39-bus New England power system. Thetest system consisted of 39 buses having 10 generators supplyingto 19 lumped loads. The simulation result shows that the proposedtechnique provides optimum load shedding. Hence, the stability ofthe power system is enhanced [27,46].

136 J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140

Kottick and Or [47] used two neural network models to solvethe forced outage of a generating unit fault in an isolated powersystem. The first neural network determined the minimal fre-quency during the forced outage of a generating unit and the sec-ond neural network predicted how many load shedding stageswould be required for the contingency. Another application ofANN in supporting the decision making of power system operatorstowards the quick stabilisation of multi-machine systems is dis-cussed in [48].

Hsu et al. [49] proposed an ANN-based load shedding techniqueto enhance the reliability of the Taiwan tie-power system. Taiwansuffered from a severe blackout in 1999 due to line tripping be-tween the southern and northern-central regions. This blackout af-fected 82.5% of Taiwanese consumers [50–52]. The Taiwan tie-power system is currently using a 15-stage load shedding tech-nique. The simulation results demonstrate that the proposed tech-nique provides the exact amount of load shedding. Due to theseadvantages, the proposed system can be used in real-time applica-tions [49]. The ANN application for alleviating overloaded lines be-fore power system separation occurs is presented in [53] andverified on the IEEE 30-bus system. Purnomo et al. [54] has pre-sented the ANN-based load shedding technique for anticipatingdrastic frequency drops in the power system.

Mitchell et al. [55] presented an ANN-based strategy for quicklyand optimally predicting the dynamic response of a power system.An efficient computational technique based on ANN to predict thesuitable strategy for setting a UFLS relay is discussed in [56]. Java-dian et al. [57] proposed an ANN-based technique for severe faultscontingencies occurred in DG-based distribution network. The pro-posed technique perform load shedding by splitting the distribu-tion network into several zones, each capable of operating inislanding mode [57].

4.1.1. Limitations of ANN in load shedding applicationsDespite the advantages that ANN has over the conventional

techniques, it has several limitations that may restrict its imple-mentation in real-time applications. Research has proven thatANN can provide satisfactory results for known (trained) casesonly. ANN fails to predict accurate results for unknown (untrained)or varying cases [58]. This means that ANN will not provide accu-rate output regarding cases that are not included in ANN training.A study dealing with this issue was carried out by testing ANNusing two configurations. In the first configuration, the ANN outputlayer had one neuron, while in the second configuration, it had sixneurons. Each ANN was tested on 22 unseen cases to predict theload shed amount. With the first ANN configuration, only 10 un-seen cases were correctly assessed. However, with the secondANN configuration, ANN correctly assessed 11 cases, 6 remainwithout a decision, and 5 were wrongly assessed. This shows thatthe performance of ANN involving unknown cases is poor and failsto provide accurate output values. One reason for this poor re-sponse may be that the training data had relatively few patternsclosely associated with unknown cases. It was also concluded thatANN proves to be a very good interpolator, but not an extrapolator.This unpredictable behavior of ANN casts doubt on its reliability[58]. A similar conclusion – that ANN fails to provide accurate re-sults under varying network situations – was drawn by Hobsonand Allen [59].

4.2. Fuzzy logic control (FLC) application in load shedding

FLC is a mathematical tool suitable for modeling a system whichis too complex and not well defined by mathematical formulation.FLC has been widely applied in almost every part of a power sys-tem. To cover the detailed application of FLC in power systems isbeyond the scope of this paper. Some of these applications include

load frequency control [60–63], unified power flow controller(UPFC) application [64,65], flexible AC transmission system(FACTS) application [66,67], and reactive power/voltage control[68,69].

Various researchers have applied fuzzy logic control for loadshedding application. A fuzzy controller has been used for intelli-gent load shedding to provide vulnerability control in a grid-con-nected power system [70]. The performance of FLC on the IEEE300-bus test system shows that it enables accurate load sheddingduring contingencies. The fuzzy logic application for avoiding volt-age collapse by shedding weak load buses is presented in [71]. Thetechnique was tested on the Ward-Hale 6-bus system and the IEEE14, 30, and 57-bus systems. The simulation results show that theproposed technique can be implemented successfully on a systemof any size.

Sallam and Khafaga [72] applied fuzzy logic control for loadshedding to obtain voltage stability in an IEEE 14-bus system. Sim-ulation results show that load shedding with the fuzzy logic con-troller successfully stabilized the system and restored the voltageto a nominal value. Another application of FLC for load sheddingis specifically to arrest dynamic voltage instability as presentedin [73]. In an islanded distribution system, the power system fre-quency is very sensitive to load variation and may cause generatoroutages or overloading if not restored quickly and properly. Tosolve this problem, a new fuzzy logic based UFLS technique forislanded mode was developed and is presented in [74]. The pro-posed technique was formulated on frequency (f), rate of changeof frequency (df/dt), and load prioritization. The technique wastested on several generator tripping and overload events. The sim-ulation results show that a fuzzy-based technique provides opti-mum load shedding and successfully restores the frequency to anominal value [74].

4.3. Adaptive neuro-fuzzy inference system (ANFIS) application in loadshedding

The ANFIS method is based on the combination of artificial neu-ral networks and fuzzy logic control. ANFIS combines the learningabilities of ANN with the fuzzy interpretation of the FLC system[75,76]. The research in ANFIS methods for application in manypower system problems has grown considerably. A few applica-tions include voltage contingency [77], power contingency [78],dynamic security assessment [79], short-term load forecasting[80], power system stabilizer [81,82], transmission line faults[83], and power quality [84].

ANFIS has also been applied for load shedding application in apower system. An application of ANFIS for intelligent load shed-ding to determine the load shed amount is presented in [70]. Thetechnique was validated on the IEEE 300-bus test system. The testresults show that the ANFIS technique provides an accurateamount of load shed and has the potential to be used in real-timeapplications. A similar application of ANFIS was applied on theMalaysian 87-bus test system for its vulnerability control [25,70].Bikas et al. [79] applied neuro-fuzzy decision tree technique on apower system for load shedding application. They consideredtwo case studies. The first case study deals with power systemoperation under stressed conditions, involving a load sheddingtechnique to prevent system voltage collapse. The second casedeals with the integration of wind power within existing powersystem [79].

4.4. Genetic algorithm (GA) application in load shedding

Genetic algorithms (GA) are the global optimization techniquefor solving non-linear, multi-objective problems introduced byJohn Henry Holland at the University of Michigan in 1975 [85].

Table 3Summary of computational intelligence technique application in load shedding.

No. Computational intelligence technique References

1 ANN application in load shedding [28,27,46–49,50–52,53–57,25,70,26,58,59]

2 Fuzzy logic application in loadshedding

[70–74,109]

3 ANFIS application in load shedding [25,70,79]4 Genetic algorithms application in load

shedding[90,31,88,89,95,91–94,110]

5 Particle swarm optimizationapplication in load shedding

[23,107,108,111]

Table 4Advantages and drawbacks of computational intelligence techniques.

No Technique Advantages Drawbacks

1 Artificialneuralnetworks(ANN)

ANN has the ability toensure an optimumamount of load shedding

ANN can providesatisfactory results forknown cases only andmay fail to predictaccurate results forunknown or varyingcases [58]

2 Fuzzy logiccontrol (FLC)

FLC can be used for loadshedding application on apower system of any size[71]

The membershipparameters of FLCrequire prior systemknowledge. Otherwise,it may fail to provideoptimum load shedding

3 Adaptiveneuro-fuzzyinferencesystem (ANFIS)

FLC parameters areoptimized by using ANN,which may lead toaccurate load shedding

It can only work withSugeno-type systems

4 Geneticalgorithms(GA)

GA is a globaloptimization techniquefor solving non-linear,multi-objective problems.GA ensures a minimumamount of load shedding

GAs take a long time todetermine the loadshedding amount. Thisrelative slowness limitstheir usage for onlineapplication [95]

5 Particle swarmoptimization(PSO)

PSO computation issimple and has the abilityto find the optimum value

PSO is easily interruptedby partial optimization

J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140 137

GA application in power system includes optimal reactive power[86] and over-current relay coordination [87].

GA also has some application in load shedding problems. Sana-ye-Pasand and Davarpanah [88] applied a genetic algorithm forload shedding applications in power systems. The database for loadshedding problems was obtained from a power flow study and wassuccessfully implemented on the IEEE 30-bus system. Another GA-based load shedding technique that considers the load sheddingfrom each bus is proposed in [89]. The GA and PSO techniques wereused to solve generator outage and line outage cases, and were val-idated on the IEEE 30-bus system. The responses of GA and PSO inall the case studies were compared. The results show that in termsof computation time, PSO is faster than GA; the minimum amountof load is shed by GA [89].

The genetic algorithms application for minimization of the loadshed amount is proposed in [90] for a single-machine infinite bus.The technique was tested by simulating the 12-month load de-mand for an optimal UFLS setting and the results compared witha conventional technique. The results indicate that the GA-basedtechnique is feasible and effective in providing optimal load shed-ding [90]. Luan et al. [91] discussed a GA-based method to deter-mine the supply restoration and optimal load shedding strategyfor distribution networks.

An effort to determine the UFLS relay setting for an isolatedpower system and a micro grid by genetic algorithm is performedin [92] and [93], respectively. GA was employed to determine theminimum load shed at each stage for the under-frequency relay.The proposed method was validated on an isolated power systemthat includes wind and diesel power generators [92], and on a mi-cro grid test system having a gas turbine, a wind turbine, and a so-lar power system [93]. Lopes et al. [31] proposed a GA-basedmethod to determine the optimal load shedding technique for con-tingencies. The proposed technique has proved to be feasible andefficient. Another application of GA for the security assessment ofa power system when subjected to a loss of K components is pre-sented in [94]. GA treated this problem as a bi-level program inwhich upper-level optimization identified a set of out-of-servicecomponents in the power system, while lower-level optimizationmodeled the reaction of the system operator during these outages.The results show the effective performance of GA in terms of solu-tion quality.

4.4.1. Limitation of genetic algorithmsThe main drawback of genetic algorithms which restricts its

implementation in real-time application is its slow response. Ithas been observed that the computation time of GAs to determinethe amount of load shed is very large. This relative slowness limitstheir usage for online application [95].

4.5. Particle swarm optimization (PSO) application in load shedding

Kennedy and Eberhart introduced the PSO technique in 1995,inspired by the social behavior of organisms such as birds flockingand fish schooling [96]. PSO has been proved as a robust and fasttechnique in solving non-linear, multi-objective problems. ThePSO technique has been widely adopted in power engineeringapplications. Most of its applications deal with economic load dis-patch (ELD) problems such as those involving the valve-point effect[97–99], the non-smooth cost function [100,101], generator con-straints [102], unit commitment [103], the unified power flow con-troller (UPFC) for the damping of power system oscillation[104,105], and load frequency control [106].

PSO has also been successfully implemented for load sheddingapplication in power systems. PSO application in an optimal loadshedding algorithm to determine the maximum loading point orcollapse point is discussed in [23]. The technique was validated

on the IEEE 14-bus system and was also compared with the GAtechnique. It was found that PSO finds the global optimum solutionmore quickly as compared to genetic algorithms [23]. A hybrid ap-proach called a particle swarm-based-simulated annealing optimi-zation technique has also been applied in an under-voltage loadshedding problem [107]. The technique provides optimal under-voltage load shedding to assist long-term voltage stability; it wasapplied on the IEEE14 and IEEE118-bus test systems. The proposedtechnique identifies the global optimum solution within a smallernumber of iterations. This PSO ability of taking only minimal timemay encourage its implementation for real-time optimal loadshedding in power systems [107]. A comprehensive learning parti-cle swarm optimization (CLPSO) has been applied to optimally par-tition the distribution system in case of main upstream loss. Ineach island, the power balance is achieved through load shedding.The proposed technique was verified on a two-test system, a 33-ra-dial bus system, and an Egyptian 66 kV, 45-bus meshed network[108].

The summary of all computational intelligence techniques withtheir references are presented in tabular form in Table 3.

5. Discussion

This paper has discussed the ability of computational intelli-gence techniques to obtain accurate load shedding within a short

138 J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140

time in emergency conditions. The review shows that computa-tional intelligence techniques are the better option for modernpower systems compared with conventional load shedding tech-niques. Computational intelligence techniques have the ability toprovide fast and optimum load shedding during contingencies toprevent power blackouts. However, each computational intelli-gence technique has certain drawbacks that restrict their imple-mentation in real-time applications. Table 4 summarizes theadvantages and disadvantages of computational intelligence tech-niques for load shedding applications in power systems.

The load shedding in a power system is a very complex and quickprocess. Faults during contingencies are unpredictable and the timerequired to perform load shedding is also very short. Load sheddingtechniques based on ANFIS may then be the most accurate optionamong the available techniques. ANFIS has the advantage ofcombining the features of fuzzy logic control and ANN, reducingtheir relative deficiencies. However, extensive research intoimproving these computational intelligence techniques is still re-quired to ensure effective implementation in real-time applications.

6. Conclusion

Recent power blackouts that have occurred around the worldmake the reliability of conventional UFLS techniques questionable.Conventional UFLS techniques are not suitable for today’s large andcomplex power systems. Computational intelligence techniqueshave the ability to deal with such modern power systems effi-ciently. This paper has presented a review of computational intel-ligence techniques as applied in load shedding and discussed therelative merits and demerits of each against the others. It can beconcluded that the implementation of computational intelligencetechniques in load shedding can reduce the possibility of blackouts,and enhance the power system’s reliability. However, furtherimprovements are still needed to make these techniques compati-ble with real-time applications.

Acknowledgements

This work is supported by the Ministry of Science, Technologyand Innovation of Malaysia (HIR-MOHE D000004-16001), the Uni-versity of Malaya and Quaid-e-Awam University of EngineeringScience & Technology Nawabshah, Sindh, Pakistan.

References

[1] USA power blackout report; 2011. <http://powerquality.eaton.com/blackouttracker> [accessed 05.09.12].

[2] El-Sadek MZ. Preventive measures for voltage collapses and voltage failures inthe Egyptian power system. Electr Power Syst Res 1998;44:203–11.

[3] Blackouts report, <https://www.allianz.com/media/responsibility/documents/position_paper_power_blackout_risks.pdf> [accessed 05.09.12].

[4] Chang L, Wu Z. Performance and reliability of electrical power grids undercascading failures. Int J Electr Power 2011;33:1410–9.

[5] Zhao X, Zhang X, He B. Study on self organized criticality of China power gridblackouts. Energ Convers Manage 2009;50:658–61.

[6] Makarov YV, Reshetov VI, Stroev A, Voropai I. Blackout prevention in theUnited States, Europe, and Russia. IEEE Proc 2005:1942–55.

[7] Andersson G, Donalek P, Farmer R, Hatziargyriou N, Kamwa I, Kundur P, et al.Causes of the 2003 major grid blackouts in North America and Europe, andrecommended means to improve system dynamic performance. IEEE T PowerSyst 2005;20:1922–8.

[8] India Blackout. <http://en.wikipedia.org/wiki/July_2012_India_blackout>[accessed 05.09.12].

[9] Pourbeik P, Kundur PS, Taylor CW. The anatomy of a power grid blackout –root causes and dynamics of recent major blackouts. IEEE Power Energy M2006;4:22–9.

[10] Wang J-W, Rong L-L. Robustness of the western United States power gridunder edge attack strategies due to cascading failures. Safe Sci2011;49:807–12.

[11] White D, Roschelle A, Peterson P, Schlissel D, Biewald B, Steinhurst W. The2003 blackout: solutions that won’t cost a Fortune. The Electr J2003;16:43–53.

[12] Power outage in world. <http://siteresources.worldbank.org/EXTESC/Resources/Approach_paper_annexes_14_October_2009.pdf> [accessed05.09.12].

[13] Mahat P, Zhe C, Bak-Jensen B. Underfrequency load shedding for an islandeddistribution system with distributed generators. IEEE T Power Deliver2010;25:911–8.

[14] Kundur P. Power system stability and control. 1st ed., United States ofAmerica; 1994.

[15] ENTSOE operation handbook. <https://www.entsoe.eu/publications/system-operations-reports/operation-handbook/> [assessed 13.05.13].

[16] Tang J, Liu J, Ponci F, Monti A. Adaptive load shedding based on combinedfrequency and voltage stability assessment using synchrophasormeasurements. IEEE T Power Syst 2013;28:2035–47.

[17] IEEE standard. IEEE guide for the application of protective relays used forabnormal frequency load shedding and restoration. IEEE Std C37117-2007;c1–43.

[18] Delfino B, Massucco S, Morini A, Scalera P, Silvestro F. Implementation andcomparison of different under frequency load-shedding schemes. Power EngSoc Summer Meet 2001;1:307–12.

[19] Lukic M, Kuzle I, Tesnjak S. An adaptive approach to setting underfrequencyload shedding relays for an isolated power system with private generation.9th Mediterranean Electrotech Conf 1998;2:1122–5.

[20] Zin AAM, Hafiz HM, Wong WK. Static and dynamic under-frequency loadshedding: a comparison. Int Conf Power Syst Technol 2004;1:941–5.

[21] Mohd Zin AA, Mohd Hafiz H, Aziz MS. A review of under-frequency loadshedding scheme on TNB system. Proc Natl Power Energy Conf 2004:170–4.

[22] IEEE guide for abnormal frequency protection for power generating plants.ANSI/IEEE Std C37106-1987;0_1.

[23] Amraee T, Mozafari B, Ranjbar AM. An improved model for optimal undervoltage load shedding: particle swarm approach. IEEE Power India Conf2006:6.

[24] Saadat H. Power system analysis. 1st ed. Singapore: McGraw-Hill; 1999.[25] Haidar AMA, Mohamed A, Al-Dabbagh M, Hussain A. Vulnerability

assessment and control of large scale interconnected power systems usingneural networks and neuro-fuzzy techniques. Power Eng Conf 2008:1–6.

[26] Haidar AMA, Mohamed A, Hussain A, Jaalam N. Artificial intelligenceapplication to Malaysian electrical power system. Exp Syst Appl2010;37:5023–31.

[27] Hooshmand R, Moazzami M. Optimal design of adaptive under frequencyload shedding using artificial neural networks in isolated power system. Int JElectr Power 2012;42:220–8.

[28] Hsu C-T, Chuang H-J, Chen C-S. Adaptive load shedding for an industrialpetroleum cogeneration system. Exp Syst Appl 2011;38:13967–74.

[29] Cheng-Ting H, Hui-Jen C, Chao-Shun C. Artificial neural network basedadaptive load shedding for an industrial cogeneration facility. IEEE Ind ApplSoc 2008:1–8.

[30] Croce F, Delfino B, Fazzini PA, Massucco S, Morini A, Silvestro F, et al.Operation and management of the electric system for industrial plants: anexpert system prototype for load-shedding operator assistance. IEEE T IndAppl 2001;37:701–8.

[31] Lopes JAP, Wong Chan W, Proenca LM. Genetic algorithms in the definition ofoptimal load shedding strategies. Int Conf Electr Power Eng 1999:154.

[32] Jeyasurya B. Artificial neural networks for on-line voltage stabilityassessment. IEEE Power Eng Soc 2000:2014–8.

[33] Aggoune M, El-Sharkawi MA, Park DC, Dambourg MJ, Marks II RJ. Preliminaryresults on using artificial neural networks for security assessment (of powersystems). Power Ind Comput Appl Conf 1989:252–8.

[34] Aggoune M, El-Sharkawa MA, Park DC, Damborg MJ, Marks II RJ. Preliminaryresults on using artificial neural networks for security assessment [of powersystems]. IEEE T Power Syst 1991;6:890–6.

[35] Aggoune M, El-Sharkawi MA, Park DC, Damborg MJ, Marks II RJ. Correction topreliminary results on using artificial neural networks for securityassessment (May 91 890–896). IEEE T Power Syst 1991;6:1324–5.

[36] Mori H. An artificial neural-net based method for estimating power systemdynamic stability index. Proc Int Forum Appl Neural Networks Power Syst1991:129–33.

[37] Sobajic DJ, Pao YH. Artificial neural-net based dynamic security assessmentfor electric power systems. IEEE T Power Syst 1989;4:220–8.

[38] Chao-Rong C, Yuan-Yin H. Synchronous machine steady-state stabilityanalysis using an artificial neural network. IEEE T Energy Convers1991;6:12–20.

[39] Chao-Rong C, Yuan-Yih H. Synchronous machine steady-state stability annlysisusing an artificial neural network. IEEE Power Eng Rev 1991;11:32–3.

[40] Aboytes F, Ramirez R. Transient stability assessment in longitudinal powersystems using artificial neural networks. IEEE T Power Syst 1996;11:2003–10.

[41] Edwards AR, Chan KW, Dunn RW, Daniels AR. Transient stability screeningusing artificial neural networks within a dynamic security assessmentsystem. IET Gener Transm Dis 1996;143:129–34.

[42] Park DC, El-Sharkawi MA, Marks II RJ, Atlas LE, Damborg MJ. Electric loadforecasting using an artificial neural network. IEEE T Power Syst1991;6:442–9.

[43] Hartana RK, Richards GG. Harmonic source monitoring and identificationusing neural networks. IEEE T Power Syst 1990;5:1098–104.

[44] Dash PK, Pradhan AK, Panda G. Application of minimal radial basis functionneural network to distance protection. IEEE T Power Deliver2001;16:68–74.

J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140 139

[45] Ebron S, Lubkeman DL, White M. A neural network approach to the detectionof incipient faults on power distribution feeders. IEEE T Power Deliver1990;5:905–14.

[46] Moazzami M, khodabakhshian A. A new optimal adaptive under frequencyload shedding using artificial neural networks. Iranian Conf on Electr Eng(ICEE) 2010:824–9.

[47] Kottick D, Or O. Neural-networks for predicting the operation of an under-frequency load shedding system. IEEE T Power Syst 1996;11:1350–8.

[48] Djukanovic M, Sobajic DJ, Pao YH. Neural net based determination ofgenerator-shedding requirements in electric power systems. IET GenerTransm Dis 1992;139:427–36.

[49] Hsu CT, Kang MS, Chen CS. Design of adaptive load shedding by artificialneural networks. IET Gener Transm Dis 2005;152:415–21.

[50] Lee CH, Hsieh SC. Lessons learned from the power outages on 29 July and 21September 1999 in Taiwan. IET Gener Transm Dis 2002;149:543–9.

[51] Wong J-J, Su C-T, Liu C-S, Chang C-L. Study on the 729 blackout in the Taiwanpower system. Int J Electr Power 2007;29:589–99.

[52] Chien-Hsing L, Shih-Chieh H. A technical review of the power outage on July29 in Taiwan. IEEE Power Eng Soc 1999;2001:1353–8.

[53] Novosel D, King RL. Using artificial neural networks for load shedding toalleviate overloaded lines. IEEE T Power Deliver 1994;9:425–33.

[54] Purnomo MH, Patria CA, Purwanto E. Adaptive load shedding of the powersystem based on neural network. TENCON Proc Comput, Commun, ControlPower Eng 2002:1778–81.

[55] Mitchell MA, Lopes JAP, Fidalgo JN, McCalley JD. Using a neural network topredict the dynamic frequency response of a power system to an under-frequency load shedding scenario. IEEE Power Eng Soc 2000:346–51.

[56] Thalassinakis EJ, Dialynas EN, Agoris D. Method combining ANNs and MonteCarlo simulation for the selection of the load shedding protection strategies inautonomous power systems. IEEE T Power Syst 2006;21:1574–82.

[57] Javadian SAM, Haghifam MR, Bathaee SMT, Fotuhi Firoozabad M. Adaptivecentralized protection scheme for distribution systems with DG using riskanalysis for protective devices placement. Int J Electr Power 2013;44:337–45.

[58] Tso SK, Zhu TX, Zeng QY, Lo KL. Investigation of extended fuzzy reasoning andneural classification for load-shedding prediction to prevent voltageinstability. Electr Power Syst Res 1997;43:81–7.

[59] Hobson E, Allen GN. Effectiveness of artificial neural networks for first swingstability determination of practical systems. IEEE T Power Syst 1994;9:1062–8.

[60] Çam E. Application of fuzzy logic for load frequency control of hydroelectricalpower plants. Energ Convers Manage 2007;48:1281–8.

[61] Çam E, Kocaarslan _I. Load frequency control in two area power systems usingfuzzy logic controller. Energ Convers Manage 2005;46:233–43.

[62] El-Sherbiny MK, El-Saady G, Yousef AM. Efficient fuzzy logic load-frequencycontroller. Energ Convers Manage 2002;43:1853–63.

[63] Kocaarslan _I, Çam E. Fuzzy logic controller in interconnected electrical powersystems for load-frequency control. Int J Electr Power 2005;27:542–9.

[64] Mishra S, Dash PK, Panda G. TS-fuzzy controller for UPFC in a multimachinepower system. IET Gener Transm Dis 2000;147:15–22.

[65] Venkatesh B, George MK, Gooi HB. Fuzzy OPF incorporating UPFC. IET GenerTransm Dis 2004;151:625–9.

[66] Dash PK, Mishra S, Panda G. Damping multimodal power system oscillationusing a hybrid fuzzy controller for series connected FACTS devices. IEEE TPower Syst 2000;15:1360–6.

[67] Dash PK, Morris S, Mishra S. Design of a nonlinear variable-gain fuzzycontroller for FACTS devices. IEEE T Contrl Syst T 2004;12:428–38.

[68] Khazali AH, Kalantar M, Khazali A. Fuzzy multi-objective reactive powerclearing considering reactive compensation sources. Energy 2011;36:3319–27.

[69] Zhang W, Liu Y. Multi-objective reactive power and voltage control based onfuzzy optimization strategy and fuzzy adaptive particle swarm. Int J ElectrPower 2008;30:525–32.

[70] Haidar AMA, Mohamed A, Hussain A. Vulnerability control of large scaleinterconnected power system using neuro-fuzzy load shedding approach. ExpSyst Appl 2010;37:3171–6.

[71] Sasikala J, Ramaswamy M. Fuzzy based load shedding strategies for avoidingvoltage collapse. Appl Soft Comput 2011;11:3179–85.

[72] Sallam AA, Khafaga AM. Fuzzy expert system using load shedding for voltageinstability control. Eng Syst Conf Power Eng 2002:125–32.

[73] Tso SK, Zhu TX, Zeng QY, Lo KL. Evaluation of load shedding to preventdynamic voltage instability based on extended fuzzy reasoning. IET GenerTransm Dis 1997;144:81–6.

[74] Mohkhlis H, Laghari JA, Bakar AHA, Karimi M. A fuzzy based under-frequencyload shedding scheme for islanded distribution network connected with DG.Int Rev Electr Eng 2012;7:4992–5000.

[75] Onat N, Ersoz S. Analysis of wind climate and wind energy potential ofregions in Turkey. Energy 2011;36:148–56.

[76] ZareNezhad B, Aminian A. Accurate prediction of sour gas hydrateequilibrium dissociation conditions by using an adaptive neuro fuzzyinference system. Energ Convers Manage 2012;57:143–7.

[77] Chaturvedi KT, Pandit M, Srivastava L, Sharma J, Bhatele RP. Hybrid fuzzy-neural network-based composite contingency ranking employing fuzzycurves for feature selection. Neurocomputing 2009;73:506–16.

[78] Chauhan S. Fast real power contingency ranking using counter propagationnetwork: feature selection by neuro-fuzzy model. Electr Power Syst Res2005;73:343–52.

[79] Bikas AK, Voumvoulakis EM, Hatziargyriou ND. Neuro-fuzzy decision trees fordynamic security control of power systems. Int Conf Intell Syst Appl PowerSyst 2009:1–6.

[80] Khotanzad A, Enwang Z, Elragal H. A neuro-fuzzy approach to short-term loadforecasting in a price-sensitive environment. IEEE T Power Syst2002;17:1273–82.

[81] Ruhua Y, Eghbali HJ, Nehrir MH. An online adaptive neuro-fuzzy powersystem stabilizer for multimachine systems. IEEE T Power Syst 2003;18:128–35.

[82] Abido MA, Abdel-Magid YL. A hybrid neuro-fuzzy power systemstabilizer for multimachine power systems. IEEE T Power Syst 1998;13:1323–30.

[83] Reddy MJ, Mohanta DK. Adaptive-neuro-fuzzy inference system approach fortransmission line fault classification and location incorporating effects ofpower swings. IET Gener Transm Dis 2008;2:235–44.

[84] Chiung-Chou L, Hong-Tzer Y. Recognizing noise-influenced power qualityevents with integrated feature extraction and neuro-fuzzy network. IEEE TPower Deliver 2009;24:2132–41.

[85] Holland John Henry. Adaptation in natural and artificial systems: anintroductory analysis with applications to biology, control, and artificialintelligence. 5th ed. United States of America: University of Michigan Press;1975.

[86] Chao-Rong C, Hang-Sheng L, Wenta T. Optimal reactive power planning usinggenetic algorithm. IEEE Int Conf Syst Man Cybernet 2006:5275–9.

[87] Cheng-Hung L, Chao-Rong C. Using genetic algorithm for overcurrent relaycoordination in industrial power system. Int Conf Intell Syst Appl Power Syst2007:1–5.

[88] Sanaye-Pasand M, Davarpanah M. A new adaptive multidimensional loadshedding scheme using genetic algorithm. Canadian Conf on Electr ComputEng 2005:1974–7.

[89] Rad BF, Abedi M. An optimal load-shedding scheme during contingencysituations using meta-heuristics algorithms with application of AHP method.Int Conf Optimiz Electr Electron Equip 2008:167–73.

[90] Chao-Rong C, Wen-Ta T, Hua-Yi C, Ching-Ying L, Chun-Ju C, Hong-Wei L.Optimal load shedding planning with genetic algorithm. IEEE Ind Appl Soc2011:1–6.

[91] Luan WP, Irving MR, Daniel JS. Genetic algorithm for supply restoration andoptimal load shedding in power system distribution networks. IET GenerTransm Dis 2002;149:145–51.

[92] Ying-Yi H, Po-Hsuang C. Genetic-based underfrequency load shedding in astand-alone power system considering fuzzy loads. IEEE T Power Deliver2012;27:87–95.

[93] Hong YY, Hsiao MC, Chang YR, Lee YD, Huang HC. Multiscenariounderfrequency load shedding in a microgrid consisting of intermittentrenewables. IEEE T Power Deliver; 2013. p. 1 doi.10.1109/tpwrd.2013.2254502.

[94] Arroyo JM, Fernández FJ. Application of a genetic algorithm to n–K powersystem security assessment. Int J Electr Power 2013;49:114–21.

[95] Al-Hasawi WM, El Naggar KM. Optimum steady-state load-sheddingscheme using genetic based algorithm. Electrotech Conf MELECON2002:605–9.

[96] Kennedy J, Eberhart R. Particle swarm optimization. IEEE Int Conf NeuralNetworks 1995:1942–8.

[97] Yuan X, Su A, Yuan Y, Nie H, Wang L. An improved PSO for dynamicload dispatch of generators with valve-point effects. Energy 2009;34:67–74.

[98] Ke M, Hong Gang W, ZhaoYang D, Kit Po W. Quantum-inspired particleswarm optimization for valve-point economic load dispatch. IEEE T PowerSyst 2010;25:215–22.

[99] Chakraborty S, Senjyu T, Yona A, Saber AY, Funabashi T. Solving economicload dispatch problem with valve-point effects using a hybrid quantummechanics inspired particle swarm optimisation. IET Gener Transm Dis2011;5:1042–52.

[100] Lu H, Sriyanyong P, Song YH, Dillon T. Experimental study of a new hybridPSO with mutation for economic dispatch with non-smooth cost function. IntJ Electr Power 2010;32:921–35.

[101] Jong-Bae P, Ki-Song L, Joong-Rin S, Lee KY. A particle swarm optimization foreconomic dispatch with nonsmooth cost functions. IEEE T Power Syst2005;20:34–42.

[102] Safari A, Shayeghi H. Iteration particle swarm optimization procedure foreconomic load dispatch with generator constraints. Exp Syst Appl2011;38:6043–8.

[103] Jacob Raglend I, Raghuveer C, Rakesh Avinash G, Padhy NP, Kothari DP.Solution to profit based unit commitment problem using particle swarmoptimization. Appl Soft Comput 2010;10:1247–56.

[104] Shayeghi H, Shayanfar HA, Jalilzadeh S, Safari A. Design of output feedbackUPFC controller for damping of electromechanical oscillations using PSO.Energ Convers Manage 2009;50:2554–61.

[105] Shayeghi H, Shayanfar HA, Jalilzadeh S, Safari A. A PSO based unified powerflow controller for damping of power system oscillations. Energ ConversManage 2009;50:2583–92.

[106] Shayeghi H, Jalili A, Shayanfar HA. Multi-stage fuzzy load frequency controlusing PSO. Energ Convers Manage 2008;49:2570–80.

[107] Sadati N, Amraee T, Ranjbar AM. A global particle swarm-based-simulatedannealing optimization technique for under-voltage load shedding problem.Appl Soft Comput 2009;9:652–7.

140 J.A. Laghari et al. / Energy Conversion and Management 75 (2013) 130–140

[108] El-Zonkoly A, Saad M, Khalil R. New algorithm based on CLPSO for controlledislanding of distribution systems. Int J Electr Power 2013;45:391–403.

[109] Calderaro V, Galdi V, Cortes-Carmona M, Palma-Behnke R. Fuzzy load-shedding strategy in distribution systems. Int Conf Intell Syst Des Appl2011:319–24.

[110] Tarafdar Hagh M, Galvani S. A multi objective genetic algorithm for weightedload shedding. In: 18th Iranian conference on electrical engineering (ICEE);2010. p. 867–73.

[111] Yinliang X, Wenxin L, Jun G. Stable multi-agent-based load sheddingalgorithm for power systems. IEEE T Power Syst 2011;26:2006–14.