performance evaluation of swarm intelligence based power system ...

PERFORMANCE EVALUATION OFSWARM INTELLIGENCE BASED POWER SYSTEMOPTIMIZATION STRATEGIESbyP. AJAY-D-VIMAL RAJUnder tbe Supervision ofDr. T.G. PALANIVELUA thesis submitted in fuljilment for the award of the degree ofDOCTOR OF PHILOSOPHYinELECTRONICS AND COMMUNICATION ENGINEERINGDEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERJNCPONDlCHERRY ENGINEERING COLLEGEPONDlCHERRY UNIVERSITYPUDUCHERRY - 605 014INDIAAPRIL 2008

BONAFIDE CERTIFlCATECertified that this thesis titled "PERFORMANCE EVALUATION OF SWARMINTELLIGENCE BASED POWER SYSTEM OPTIMIZATION STRATEGIES is thebonafide work of Mr. P. AJAY-D-VIMAL RAJ who canied out the research work undermy supervision. Certified further. that to the best of my knowledge, the work reported hereindoes not form part of any other thesis or dissertation on the basis of which a degree or awardwas conferred on an earlier occasion on this or any other candidate.PrincipalPondicheny Engineering CollegePuducheny - 605 014INDIA.

DECLARATION1 hereby declare that the thesis titled "PERFORMANCE EVALUATION OFSWARM INTELLIGENCE BASED POWER SYSTEM OPTIMIZATIONSTRATEGIES" submitted to the Pondicheny University in fUlfillment of the reqiurementsfor the award of the degree of DOCTOR OF PHILOSOPHY in Electronics andCommunication Engineering, is a record of the original research work done by me under thesupervision of Dr.T.G.Palanivelu, Professor. Department of Electronics and CommunicationEngineering, Pondicheny Engineering College and that the work has not been submittedeither in whole or in part for any other degree or at any other university.I+Signature(b?: T.G. Palanivelu)(P. Ajay-D-Vimal Raj)

ABSTRACTImplementation of Particle Swann Optimization (PSO) applications in solving theoptimization problems in the field of electric power system is proposed. PSO is a powerfultool for optimizing multidimensional discontinuous nonlinearities.The optimization problems in power system like economic load dispatch, unitcommitment, real power loss minimization, power quality improvement using distributedgeneration and minimization of energy capacity of a dynamic voltage restorer are identifiedand analyzed.PSO algorithm is applied to determine the optimal power generation of each unit thatoperates for a specific period. This minimizes the total generation cost by considering thegenerator output as the control variable. The simulations are carried out for (i) a three-unitthermal plant system (ii) a six-unit thermal plant system and (iii) a three-unit thermal plantsystem in which one unit is a combined cycle co-generation plant.Due to strict regulations on environmental protection, the conventional operation atminimum fuel cost cannot be the only basis for dispatching electric power. Therefore, thecombination of economic and emission dispatch (CEED) is required which leads to a dualobjective problem. The purpose of combined economic and emission dispatch problem is toidentify the optimal amount of generated power for the generating units in the system byminimizing the fuel cost and emission level simultaneously subjected to various systemconstraints. The multi objective CEED problem that includes both generation cost andemission cost is solved by swarm intelligence. PSO is applied to the dual objective CEEDproblem by using a price penalty factor approach to form a single objective. The ParticleSwarm Optimization (PSO) based algorithm is developed for finding out the optimal powerdispatch in CEED environment of thermal units while satisfying the constraints such asgenerator capacity limits, power balance and the line flow limits. Using this algorithm, theglobal optimal generation value is calculated for an IEEE - 30 bus system.

In practical systems, the operating range of all the on-line units is restricted by theirramp rate limits and prohibited operating zones due to physical operational constraints. Theprohibited operating zone of an unit divide the operating range between its minimum andmaximum generation limits into several disjoint convex sub regions. Hence conventionalmethods cannot be directly applied to solve economic load dispatch problem with prohibitedoperating zones. Therefore, a hybrid PSO algorithm is used to solve the economic loaddispatch problem of generating units with prohibited operating zones and ramp rate limits.The general system constraints such as system demand, balance of power with network lossesand generating capacity limits are also incorporated. The hybrid method incorporates theGenetic Algorithm (GA) to explore the high performance region in solution space and PSOalgorithm to exploit the solution space for locating the optimal solution. Thus, the GA guidesPSO for better performance in the complex solution space. The proposed algorithm is used tosolve a 6, a 15 and a 40 unit test system.Unit commitment is the problem of determining the schedule of generating unitswithin a power system subjected to operating constraints. A hybrid GA-PSO algorithm isdeveloped in this work for solving the unit commitment problem. The algorithm involves anexhaustive study of all the possible combinations of units to meet the load demand and thecombinations corresponding to minimum production cost. GA is used to obtain the ONIOFFstatus of various generators with an objective function to have generation greater than thedemand. The initial solution obtained from GA is applied to PSO based unit commitmentalgorithm to obtain minimum production cost. The hybrid genetic based unit commitmentalgorithm is tested on a 10-unit system.The losses that occur in a power system have to be minimized m order to enhance itsoverall performance. Therefore, in this proposed work a PSO based algorithm is attempted tominimize the real power losses, with a view to improve the voltage stability of the system.The proposed PSO based algorithm aims at finding optimum settings of Automatic VoltageRegulator, On Load Tap Changer values and requires minimum number of Reactive PowerCompensation Equipments subject to equality and inequality constraints. The voltagestability assessment is performed using a line voltage stability index. The particle swarmoptimization technique is used for different systems such as Standard-5, IEEE-14, IEEE-30.IEEE-57, IEEE-I 18 bus systems, Indian Utility systems such as Neyveli Thermal PowerStation bus system and Puducheny bus system.

The introduction of Distributed Genmtion @G) in a distribution system offersseveral benefits to utilities, customers, and society. A general approach is applied to assessthe major technical benefits to improve power quality issues such as voltage profileimprovement and line loss reduction. The proposed PSO based algorithm determines theoptimal value of the DG capacity to be connected with the existing system therebymaximizing the power quality by reducing the line losses and increasing the voltage profile ofthe various buses. The line voltage stability index obtained by performing a Newton-Raphson (NR) load flow solution is used to validate the significance of the PSO basedapproach. The line index finds out the increase in maximum loadability of the system afterusing the proposed method.The growing interest in power quality has led to a variety of devices designed formitigating power disturbances. primarily voltage sags. Among several devices. a DynamicVoltage Restorer (DVR) is a novel custom power device proposed to compensate for voltagedisturbances in a distribution system. The compensation capability of DVR depends primarilyon the maximum voltage injection ability and the amount of stored energy available withinthe restorer. A novel PSO based phase advancement compensation strategy is proposed inthis work for optimizing the energy storage capacity of the DVR in order to enhance thevoltage restoration property of the device. The novel proposed algorithm is tested for asample three phase system for various levels of sag in a particular phase. The proposedalgorithm identifies the required value of phase advancement angle wmesponding to aminimum power injection from the energy storage element such as a capacitor or a battery.The results of swarm intelligence and its hybrid based optimization techniques forvarious power system problems are compared with the conventional, GA and PSO basedmethods.

ACKNOWLEDGEMENTI express my deep sense of gratitude to my esteemed research supervisorDr.T.G. Palanivelu, Principal. Pondicheny Engineering College, Puducheny for his parentalcare and scholarly guidance. I also thank him for his constant encouragement, painstakingefforts and patience throughout the course of this work without which the successfulcompletion of this work would not have been possible. I am deeply indebted to him forgiving me this opportunity to pursue my research work under his esteemed gu~dance.I am indebted to Dr. P. Dananjayan, Professor and Head, Department of Electronicsand Communication Engineering, Pondicherry Engineering College, Puducherry forproviding me the help and facilities required to pursue my research.My sincere thanks to the doctoral committee members Dr. K. Manivannan.Professor, Department of Electrical and Electronics Engineering, Pondicherry EngineeringCollege, Puducheny, Dr. K.M. Tamizhmani, Professor, Ramanujan School of Mathematicsand Computer Science, Pondicheny University, Puducherry for their valuahle suggestionsand help.With immense pleasure, I express my profound sense of gratitude toShri. N. Kesavan, Founder and Chairman, Sbri. M. Dhanasekaran, Managing Director,Shri. S. V. Sugumaran, Vice-Chairman, Dr. V.S.K. Venkatachalapathy, Principal. SriManakula Vinayagar Engineering College, Puducheny for inducing me to embark on thisprogramme.I also place on record my deep sense of gratitude to Dr. R Gnanadass, AssistantProfessor, Department of Electrical and Electronics Engineering, Pondichemy EngineeringCollege, Dr. M. Sudhakaran, Assistant Professor, Department of Electrical and ElectronicsEngineering, Pondicheny Engineering College, Dr. S. Jeevananthan, Senior Lecturer,Department of Electrical and Electronics Engineering, Pondicheny Engineering College and

Er. S. Sentbil Kumar, Assistant Professor, Department of Electrical and ElectronicsEngineering, Sri Manakula Vinayagar Engineering College, Puduchemy, for extending moralsupport and the technical d~scussions as and when required during this work.My sincere thanks are also due to Dr. R Balasubramauian. NTPC Chair Professor,Centre for Energy Studies, Indian Institute of Technology, New Delhi. Er.S.Ravicbandaran.Executive Engineer, Main Load Dispatch Centre, Tamil Nadu Electricity Board. Chennai,Er. S. Sbankar, Executive Engineer, Pondicheny Electricity Department, Puducherry fortheir valuable suggestions.1 wish to thank my student Mr. S. Prasadh Kannah for his timely help and supportduring the work. I am thankful to Mrs. K. Guejalatchoumy and S. Dbarmendra for doingthe grammatical correction in thls workI immensely thank the faculty members and the non-teaching staff of the Departmentof Electrical and Electronics Engineering for their valuable help.My special, sincere, heartfelt grat~tude and indebtness to my sister. brothers-in-law.brother, sisters-in-law for their sincere prayers, constant encouragement and blessings.Last, but not the least, 1 am extremely grateful to my parents Er. A. Perianayagamand Mrs. P. Honorine. I would never have got this far without their unlimited dedication.support and love throughout so many years. I would like to express my profound gratitude tomy father-in-law Er. Marie Joseph Bala Baskar. I am also in debt to my wife.Mrs. Mary Josephine Cecilia Shobana and my loving son Master Sam Surya Ajay fortheir support and tolerance during many stressful periods along this enterprise.research.I bow my head and thank the Almighty for showering His blessings throughout my

CONTENTSCHAPTERTITLEPAGE No.ABSTRACTACKNOWLEDGEMENTCONTENTSLlST OF TABLESLlST OF FIGURESLlST OF ABBREVlATlONSLIST OF SYMBOLSivviiixxivxixxxixxii1. INTRODUCTION1 .I. PREAMBIE1.2. REVIEW OF TRADITIONAL STRATEGIES1.2.1. Linear and Quadratic Programming Methods1.2.2. Nonlinear Programming Methods1.2.3. Integer and Mixed-Integer Programming Methods1.2.4. Dynamic Programming Methods1.3. LITERATURE REVIEW1.3.1. Economic Load Dispatch Problems1.3.1 .I. Economic Load Dispatch1.3.1.2. CEEDI .3.1.3. ELD with POZ1.3.2. Unit Commitment1.3.3. Real Power Loss Minimization and VoltageStability Enhancement1.3.4. DG for Power Quality Improvement1.3.5. Energy Optimized DVR1.4. OBJECTIVES OF THE THESIS1.5. ORGANIZATION OF THE THESIS

CHAPTER TITLE PAGE No.2. PSO BASED ECONOMIC LOAD DISPATCH 16PROBLEMS2.1. INTRODUCTION 162.2. ECONOMIC LOAD DISPATCH PROBLEM 172.2.1. Problem Description 172.2.2. Objective Function 172.2.3. Features of Particle Swarm Optimization 192.2.4. Implementation of PSO for ELD solution 242.2.4.1. Representation of an individual string 252.2.4.2. Evaluation Function 25, 2.2.5. Algorithm of the Proposed Method 262.2.6. Numerical Example. Simulation Resultsand Analysis 272.2.6.1. Test Casel: Three-tinit Thermal System 302.2.6.2. Test Case2: Three-Unit System with CCCP 332.2.6.3. Test Case3: Six-llnit Thermal System 342.3. COMBlNED ECONOMIC EMlSSlON DISPATCHPROBLEM 352.3.1. Problem Description 352.3.2. Objective Function 362.3.3. Step-by-step algorithm 382.3.4. Simulation Results 402.4. ECONOMIC DISPATCH PROBLEM WITHPROHIBITED OPERATING ZONES 452.4.1. Problem Description 452.4.2. Objective Function 462.4.3. Features of Genetic Algorithms 492.4.3.1. Components of Genetic Algorithms 492.4.4. GA and PSO Combined Hybrid Method 512.4.4.1. Description of the Proposed Method 51

CHAPTER TITLE PAGE No.2.4.4.2. Evaluation Function2.4.4.3. Application of GA-PSO Algorithm2.4.5. Simulation Results2.4.5.1. Six-Unit System2.4.5.2. Fifteen-Unit System2.4.5.3. Forty-Unit System2.5. CONCLlJSlON3. INTEGRATED GA-PSO BASEDlJNlT COMMITMENT3.1. INTRODUCTION3.2. PROBLEM DESCRIPTION3.2.1. Objective Function3.3. PROPOSED GA-PSO BASED METHOD3.3.1. Implementation of GA for UC Solution3.3.1.1. Coding of Solu~ion3.3.1.2. Fitness Function3.3.1.3. Crossover3.3.1.4. Mutation3.3.2. Implementation PSO for ELI) Solution3.3.3. Step-by-step Algorithm3.4. SIMULATION RESULTS3.5. CONCLUSION4. PSO BASED REAL POWER LOSS MINIMIZATIONAND VOLTAGE STABILITY ENHANCEMENT4.1. n\lTRODUCTION4.2. PROBLEM DESCRIPTION4.2.1. Objective Function4.3. VOLTAGE STABILITY ASSESSMENT4.3.1. Line Voltage Stability Index

CHAPTER TITLE PAGE No.4.4. ALGORITHM OF THE PROPOSED MEMOD4.5. SIMULATION RESULTS4.5.1. Standard -5 Bus System4.5.2. Standard IEEE Systems4.5.3. Indian Utility Systems4.6. CONCLUSION5. SWARM INTELLIGENCE BASED OPTlMlZATlONOF DlSTRlBUTED GENERATIONCAPACITY FOR POWER QUALITY IMPROVEMENT5.1. INTRODUCTION5.2. APPROACH TO QUANTIFY THE BENEFITS OF DG5.2.1. Voltage Pn~file Improvement lndex5.2.2. Line Loss Reduction lndex5.2.3. Line Voltage Stability lndex5.3. PROBLEM DESCRIPTION5.3.1. Objective Function5.4. ALGORITHM OF THE PROPOSED METHOD5.5. SIMULATION RESII1,TS5.5.1. Results of Improvement in Voltage Profile andLine Loss Reduction5.5.2. Comparison of Line Voltage Stability lndex of5.6. CONCLUSIONConventional and Proposed Methods6. PARTICLE SWARM OPTIMIZATION BASEDDYNAMIC VOLTAGE RESTORER6.1. INTRODUCTION6.2. OVERVIEW OF A DVR6.3. EXISTING DVR STRATEGIES6.3.1. In-Phase Voltage Injection Technique

PACE No.6.3.2. Phase Advance Compensation Technique6.3.2.1. DVR Power Flow6.4. PROBLEM FORMULATION6.4.1. Objective Function6.5. ALGORITHM OF THE PROPOSED METHOD6.6. SIMULATION RESULTS6.7. CONCLUSION7. CONCLUSION7.1. RESULTS7.2. FUTURE WORKAPPENDICESAPPENDIX-AAPPENDIX-BAPPENDIX

TABLE No.LIST OF TABLESTITLEPAGE No.2.1.Parameters used in PSO method(3.6 and CCCP unit systems)Optimal scheduling of generators neglecting losses(3-unit system)Solution of different methods neglecting losses(3-unit system)Optimal scheduling of generators including losses(3-unit system)Solution of different methods including losses(3-unit system)Optimal scheduling of generators including CCCP(3-unit system)Solution of different methods including CCCP(3-unit system)Optimal scheduling of generators(6-unit system)Solution of different methods(6-unit system)Parameters used in PSO method(IEEE-30 bus system)ELD results obtained by various methods(IEEE-30 bus system)Minimum power dispatch results by various methods(IEEE-30 bus system)Line flows with line flow constraints(IEEE-30 bus system)CEED results(IEEE-30 bus system)

TABLE No.2.15.2.16.2.17.2.18.2.19.2.20.2.21.3.1.3.2.3.3.3.4.3.5.3.6.TITLEParameters used in GA-PSO method(6-unit, 15-unit and 40-unit systems)Optimal generator dispatch solution by various methods(6-unit system)Comparison of solution quality(6-unit system)Optimal generator dispatch solution by various methods(15-unit system)Comparison of solution quality(15-unit system)Test results of the proposed approach(40-unit system)Comparison of solution quality(40-unit system)Parameters used in GA-PSO method(10-unit system)Cost coeff~cients(10-unit system)Daily generation(I 0-unit system)UC schedule obtained using initial population(10-unit system)Simulation results of initial population(I 0-unit system)UC schedule obtained using final population(I 0-unit system)Simulation results of final population(10-unit system)Comparison of solution qualityParameters used in PSO method(Standard-5, IEEE-14, -30, -57, -118. Indian utility -23and -17 bus systems)xvPAGE No.

TABLE No. TITLE PAGE No.Comparison of real power losses with different methods 85Optimal control variables(IEEE-14 bus system)Optimal control variables(IEEE-30 and -57 bus systems)Optimal control variables(IU-NTPS-23 bus system)Solution of stability assessment using line voltage stability index(IEEE-14 bus and IU-NTPS-23 bus systems) 88Parameters used in PSO method(IEEE-30 bus system) 98Voltage pmfile improvement results for a DG rating 0.2 p.u. 99Voltage profile improvement results for a DG rating 0.3 p.u. 99Line loss reduction results for a DG rating 0.2 p.u. 100Line loss reduction results for a DG rating 0.3 p.u. 101Comparison of the conventional Newon-Raphson andProposed method in computing line voltage sstability index 102Detection of critical lines using different methods 102Solution of improvement in system loadability 103Parameters used in PSO method(single phase sag) 115Comparison of results with PAC and In-Phase injection scheme 116Line data(IEEE-30 bus system)Bus data and load flow results(IEEE-30 bus system)Generator cost and emission coefficients(IEEE -30 bus system)Transformer tap setting data(IEEE-30 bus system)Shunt capacitor data(IEEE-30 bus system)

xviiTABLE No.TITLEPAGE No.A.6.B.1.B.2.53.3.C.I.C.2.C.3.U.1.F.I.F.2.G.1.G.2.G.3.G.4.H.I.H.2.Generalized loss coeficients(IEEE-30 bus system)Generating unit capacity and coefficients(6-unit system)Ramp rate limits and prohibited operating zones(6-unit system)Generalized loss coefficients(6-unit system)Generating unit coefftcients with ramp rate limits(15-unit system)Prohibited operating zones of generating units(15-unit system)Generalized loss coefficients(15-unit system)Generating unit coefficients with ramp rate limits(40-unit system)Line data(Standrad-5 bus system)Bus data(Standrad-5 bus system)Line data(IEEE-14 bus system)Bus data and load flow results(IEEE-14 bus system)Transformer tap setting data(IEEE-14 bus system)Shunt capacitor data(IEEE-14 bus system)Line data(IEEE-57 bus system)Bus data and load flow results(IEEE-57 bus system)

TABLE No.H.3.H.4.TITLETransformer tap setting data(IEEE-57 bus system)Shunt capacitor data(IEEE-57 bus system)Sites and location of different buses(IU-NTPS-23 bus system)Line data(IU-NTPS-23 bus system)Bus data(IU-NTPS-23 bus system)Line data(IU-Puducherry-17 bus system)Bus data(IU-Puducheny-17 bus system)xviiiPAGE No.

LIST OF FIGURESFIGURE No. TITLE PAGE No.Concept of modification of a searching point by PSOSearching concept with particles in a solution space by PSOFlowchart of PSO methodFuel cost characteristics of CCCP systemPSO based ELD convergence characteristics(3-unit system)Reliability characteristics evaluation(3-unit system)PSO-based ELD convergence characteristics(IEEE-30 bus system)Best generator settings of PSO-based CEED method(IEEE-30 bus system)PSO-based CEED convergence characteristics(IEEE-30 bus system)Three possible operating conditions ofa generating unitGA-PSO convergence characteristics(15-unit system)Window crossoverGA-PSO convergence characteristics(I 0-unit system)Variation of voltage profile for different optimization methodsPSO based convergence characteristics(IEEE-I 18 bus system)Voltage prafile improvement results with different DG ratingsSchematic block diagram of power distribution systemcompensated by a DVRPhasor diagram of power distribution system during sag

XXFIGURE No.PAGE No.6.3. Convergence characteristics of PSO-PAC scheme(Single-phase sag)One line diagam(IEEE-30 bus system)One line diagram of a typical transmission systemOne line diagram(Standard-5 bus system)One line diagram(IEEE-14 bus system)One line diagam(IU-NTPS-23 bus system)One line diagram(IU-Puducheny-17 bus system

LIST OF ABBREVIATIONSPSOParticle Swarm OptimizationGenetic AlgorithmELDCCCPPOZCEEDG A-PSOOPFUCDGDVRNTPSVPllLLRlIUPACLPNLPIPDPQPSCBPS0RCPSOp.u.AVROLTCRPCELVSlGEconomic Load DispatchCombined Cycle Cogeneration PlantProhibited Operating ZoneCombined Economic Emission DispatchGenetic Algorithm--Particle Swarm OptimizationOptimal Power FlowUnit CommitmentDistributed GenerationDynamic Voltage RestorerNeyveli Thermal Power StationVoltage Profile Improvement IndexLine Loss Reduction IndexIndian UtilityPhase Advance CompensationLinear ProgrammingNun Linear ProgrammingInterior PointDynamic ProgrammingQuadratic ProgrammingSynchronous CompensatorBinary Particle Swarm OptimimtionReal-coded Particle Swarm Optimizationper unitAutonlatic Voltage RegulatorOn Load Tap ChangerReactive Power Compensation EquipmentLine Voltage Stability IndexGenerator

LIST OF SYMBOLSCI. c2s,Lpbest,gbest))Cost coefficient of generator ($h4Wh2)Cost coefficient of generator (Sh4Wh)Cost coefficient of generatorEmission coefficient of generator ( kd~~h')Emission coefficient of generator (kgA4Wh)Emission coefficient of generatorTotal operating cost ($/h or Rsh)Price penalty factor ($/kg)Modified price penalty factor ($/kg)Total fuel cost of generation ($ih or Rsh)Fuel cost function of i' generator ($/h or Rsh)Real power generation of i" generator (MW)Number of generators connected in the networkDependent unit power output (MW)Total load of the system (MW)Transmission loss of the system (MW)Real power injections at m" and n" buses (MW)Generalized loss coefficients (MW ')Minimum value of real power allowed at generator i (MW)Maximum value of real power allowed at generator i (MW)Population sizeNumber of members in a particleVelocity of individual i at iteration kPointer of iterationsWeighing factorAcceleration coefficientsRandom numbers between 0 and 1Random numbers between 0 and 1Current position of individual i at iteration kBest position of individual iBest position of the group

Initial weightFinal weightMaximum iteration numberiterFFCECP,(23P,"='QY1Pc.sl.mp:;L",Ivl6Maximum velocity limitMinimum velocity limitCurrent iteration numberOptimal cost of generation (%/h or Rsh)Fuel wst ($h)Emission cost (kg/h)Calculated real power for PQ bus i (MW)Calculated reactive power for PQ bus i (MW)Specified real power for PQ bus i (MW)Specified reactive power for PQ bus i (MW)Calculated real power of PV bus m (MW)Specified real power of PV bus m (MW)Voltage magnitude of buses (per unit)Phase angle of buses (degrees)Minimum value of voltage at bus i (per unit)Maximum value of voltage at bus i (per unit)Numher of branches or transmission linesCalculated line flow of each transmission line (MVA)Rated line flow of each transmission line (MVA)Power generation of unit i at previous hour (MW)Ramp rate limit of unit i as power generation increases (MWlh)DR,-P,""-P,'-zy: 3 y:F,PPbCRamp rate limit of unit i as power generation decreases (MWlh)Effective lower limit of ith unit with ramp rate constraint (MW)Effective upper limit of i~ unit with ramp rate constraint (MW)number of prohibited zones of a unitLower and upper limits of z prohibited zones of unit i (MW)Generation cost function ($/h)Power balance constraint

1, (1)F,@'I(t)s, (t)NhPo(t)Pit)Tup.1Tm.8G,.,T,,n.,XYLoss,QmlnQmarI>S,pmQmVKv mv,vpwllx,vp wornL,K,NLLwmLLWOIDGhVZMaximum generation cost among the individuals in the initialpopulation (S/h)Minimum generation cost among the individuals in the initialpopulation (Sh)Commitment status of i" generator at how 1Fuel cost of i" generator at hour t (Sh)Start-up cost of i" unit at hour t (Oh)Total schedule time (24 h)Load demand at hour t (MW)Real power produced by i" generator at how t (MW)Minimum uptime i" generator (h)Time duration for which i" generator is continously ON (h)Minimum down time of generator i (h)Time duration for which i* generator is continously OFF (h)Continuous variables (AVR values)Discrete variables (OLTC and SC values)Power loss (PI) at branch i (MW)Minimum limit of reactive power at a node (MVAr)Maximum limit of reactive power at a node (MVAr)Line stability index at iIh line or branchReal power at the receiving end (per unit)Reactive power at the receiving end (per unit)Voltage magnitude at the sending end (per unit)Voltage magnitude at the receiving end (per unit)Voltage magnitude at bus i (per unit)Voltage profile of the system with DG (per unit)Voltage profile of the system without I)G (per unit)Load at bus i (per unit)Weighing factor for load at bus i (per unit)Total number of load busesTotal line losses in the system with DG (per unit)Total line losses in the system without DG (per unit)Penalty factor of bus voltagesBalaoced output voltage (per unit)

I0v ISubscript jpdWm*=Balanced load cumnt @er unit)Load power factor angleSource side voltage (per unit)j* phase (j = 1,2,3)Real power supplied by DVR using in-phase voltage injectiontechnique (KW)Reactive power supplied by DVR using in-phase voltageinjection technique (KVAr)Load voltage phase advancement angle (degrees)lnput power from the source (KW)Load power (KW)Real power supplied by DVR (KW)lnput reactive power from the source (KVAr)Load reactive power (KVAr)Reactive power supplied by DVR (KVAr)Optimum value of load voltage phase advancement angle forminimum power operation (degrees)Complex bus power at the sending end (per unit)Total real power at the sending end bus (per unit)Total reactive power at sending end bus (per unit)Total reactive power at receiving end bus (per unit)Total reactive power at receiving end bus (per unit)Voltage at the sending end (per unit)Voltage at the receiving end (per unit)Voltage angle (degree)Current at the receiving end (per unit)Transmission line constantsConstant angles of transmission linesSeries line impedance magnitude of transmission lines (ohms)Total line charging susceptance (mho)Propagation constantLength of a transmission line (km)Price penalty factor associated with the last unit ($kg)Price penalty factor associated with the current unit ($kg)

Maximum power associated with the last unit (MW)Maximum power associated with the current unit (MW)

programming, mixed integer pmgramming and dynamic programming (DP). Thissection attempts to review the basic concepts of these techniques.1.2.1. Linear and Quadratic Programming MethodsLinear programming (LP) methods have linear objective functions andconstraints 17-91. These methods basically fall into two categories: simplex andinteger programming (IP) [1&17]. The main advantage of simplex method is its highcomputational efficiency. But the disadvantage is that number of iterations growsexponentially with problem size. This disadvantage can be overcome by IP methods.IP methods do not step from one comer point to the next in the manner of simplexalgorithm, but rather stay within the interior of the constrained region andprogressively move to the optimal point. Both the simplex and IP methods can beextended to have a linear and quadratic objective function when the constraints arelinear. Such methods are called quadratic programming (QP) [IS-191. LP has beenused in various power system applications such as optimal power flow [S], load flow191, reactive power planning [20], and active and reactive power dispatch [21-221.1.2.2. Nonlinear Programming MethodsIn most of the NLP methods, the approach is to start from initial conditionsand determine the 'descent direction' in which the value of objective functiondecreases for a minimization problem. A large number of NLP methods are availablethat are distinguishable by their definition and step length. Quasi-Newton method 1231that attempts to build up an approximation to Hessian matrix exhibits powerfidconvergence. If the coefficients of Hessian matrix are available analytically, Newtonmethod [24] can be applied. Some of the most successful methods in use today arebased on applying QP to solve a local optimization in a nonlinear problem. IPmethods originally developed for LP can be applied to QP and NLP problems. NLPhas been applied to solve optimal power flow 1251 and hydrothermal scheduling [26]problems.

1.23. Integer and Mixed-Integer Programming MetbodsIn cases where the independent variables can take only integer values, suchproblems are called integer programming. When some of the variables are continuous,the problem is called mixed integer programming. Mainly two approaches, namely'branch and bound' and 'cutting plane methods', have been used to solve integerproblems using mathematical programming techniques [23]. The size and complexityof integer and mixed-integer programmes that can be solved in practice depends onthe structure of the problem. Integerlmixed integer programming have been applied tovarious areas of power systems such as optimal reactive power planning [27], powersystem planning [28-291, unit commitment [30] and generation scheduling [31].1.2.4. Dynamic Programming MethodsDynamic programming (DP) based on the principle of optimality states that asubpolicy of an optimal policy must in itself be an optimal subpolicy. DP is a verypowerful technique, but it suffers from the curse of dimensionality [32]. DP has beenapplied to various areas of power systems such as reactive power control [33],transmission planning [34] and unit commitment (351.The main advantage of the intelligence based methods is that it avoids thecomplexities in the formulation of mathematical model for the power systemoptimization. However, the shortcoming of these methods is generally associated withthe required excessive computational resources. With the advent of fast processorswith large memory, these methods appear to he promising in the future.1.3. LITERATURE REVIEWThey are reviewed in a systematic way in the following sections.

1.3.1. ECONOMIC LOAD DISPATCH PROBLEMS13.1.1. Economic Load DispatchThe classical lambda iteration method has been used to solve the ELDproblem. This method utilizes an equal incremental cost criterion for systems withouttransmission losses and the penalty factors using B, matrix for systems withtransmission losses. Other methods such as gradient, Newton, linear programmingand interior pint have also been applied to solve the ELD problems [41].Zwe-Lee Gaing [42] has proposed a particle swm optimization (PSO)method for solving the economic dispatch (ED) problem in power systems. Thismethod made use of PSO for its global search capab~lity to allocate optimum loadingof each generator. The test results of three different systems have been compared withthat of GA-based approach.Jayabarathi et al. [43] have adopted a particle swarm optimization techniquefor solving the various types of economic dispatch problems. The test results of thesample systems have been compared with that of other evolutionary computingtechniques.1.3.1.2. Combined Economic Emission DispatchTalaq et al. [44] have formulated an optimal power flow problem withemission constraints where the main objective was to minimize the fuel cost and thetotal emission over a wide time period of different intervals and system demands. Thetest results of standard 5-bus and IEEE-30 bus systems display a trade-off relationshipbetween fuel cost and emission.Wong et al. [45] have developed an efficient and reliable evolutionaryprogramming-basedalgorithm for solving the environmentally constrained economicdispatch (ECED) problem. This method made use of acceleration techniques in orderto enhance the speed and robustness of the algorithm.

Venkatesh et al. [46] have built an EP algorithm to solve the CEED pmblemwith line flow constraints. The line flows in MVA have been computed directly fromthe Newto=Raphson method. A novel modified price penalty factor has beenintroduced to find the exact economic emission fuel cost with respect to the loaddemand. The test results of IEEE-14, -30 and -1 18 bus systems have been comparedwith that of other evolutionary computing techniques.Abido [47] has derived a Pareto-based multiobjective evolutionary algorithm(MOEA) for solving an environmental/economic electric power dispatch pmblem.This fuzzy-based hierarchical clustering technique has been implemented in order toobtain the best solution. The test results of an IEEE-30 bus system have beencompared with that of other traditional multiobjective optimization techniques.1.3.1.3. Economic Load Dispatch with Prohibited Operating ZonesWalters et al. 1481 have developed a genetic algorithm to solve the economicdispatch problem with valve-point effects. This algorithm has utilized payoffinformation of the candidate solutions to evaluate their optimality. The test results ofthree units system have been compared with that of dynamic programming method.Wong et al. [49] have built an incremental genetic algorithm based approachfor the determination of global or near-global optimum solution. Another techniquethat incorporates both incremental genetic theory and simulated annealing has servedto determine the economic loadings of 13 generators in a practical power system withthe effects of valve-point loading and ramping characteristics. The test results havebeen found to yield better results when compared with that of simulated annealingbased method.Chen et al. [50] have presented a GA-based method that uses the incrementalcost of encoded parameter of the system for solving the ED problem taking intoaccount the network losses, ramp rate limits, valve-point zone and prohibitedoperating zone. The numerical results of the method for a large scale 40-unit systemhave been compared with that of lambda-iteration method.

Fung et al. [51] have formulated an integrated parallel genetic algorithmincorporating tabu search (TS) and simulated annealing for solving the ED problem.The parallel computing platform has been based on a network of interconnectedpersonal computers (PCs) using TCPAP socket communication facilities. The testresults of a practical power system have been obtained to compute the optimal loadingof 13 generators.El-Gallad et al. [52] have adapted a PSO technique to solve the traditionaleconomic dispatch problem. The objective function has been formulated as acombination of piecewise quadratic cost functions with nondifferential regions,instead of adopting a single convex function for each generating unit. This innovationhas served to incorporate practical operating conditions, such as valve-point effectsand fuel types. The effectiveness of the algorithm has been tested on a three unitsystem and the results have been compared with that of a numerical method.El-Gallad et al. [53] have added new constraints to the problem by introducingsystem spinning reserve and generator prohibited operating zones. In this formulation,they have included the same constraints but considered a single convex cost function1521. The test results of a 15-unit system in which four units with prohibited operatingzones have been compared with for both conventional method and the Hopfield neuralnetwork.Lai et al. 1541 have applied PSO to solve economic dispatch (ED) of units withnonsmooth input-output characteristic functions. The test results of an IEEE-30 bussystem with six generating units have been compared with that of evolutionaryprogramming (EP).Victoire et al. have extended Gaing's research by forming a hybrid optimizerto tackle the same problem [55]. They have used sequential quadratic programming tofine-tune the PSO search in finding the optimal solution. The feasibility has beenillustrated by conducting case studies on a 10-unit system with valve-point effects forthree different load-demand patterns and the results have been compared with thatobtained using the EP-SQP method.

13.2. UNIT COMMITMENTSheble et al. [56] have presented a genetic-based unit commitment (UC)scheduling algorithm. It has made use of GA with domain specific mutation operatorsfor finding good unit commitment schedules. The test results of three different electricutilities have been compared with that of Lagrangian relaxation UC method.Bakirtzis et al. [57] have developed a genetic algorithm that uses differentquality function techniques to solve the unit commitment problem. The test results upto 100 generator units have been compared with that of dynamic programming andLagrangian relaxation methods.Swarup et al. [58] have employed a new solution methodology to the UCproblem using genetic algorithm. The strategy has been found to be efficient andserve to handle larger size UC problems.Zwe-Lee Gaing 1591 has built an integrated approach of discrete binaryparticle swarm optimization (BPSO) with the lambda-iteration method for solving theUC problem. It has been solved as two subproblems using BPS0 method forminimization of the transition cost. The economic dispatch problem has been solvedby lambda-iteration method for the minimization of the production cost. Thefeasibility of the method has been demonstrated on a 10- and a 26-unit systems, andthe test results have been compared with that of GA method.Zhao et al. [60] have presented an improved particle swarm optimization(IPSO) algorithm for power system UC problem. It has adopted an orthogonal designin order to generate the initial population that are scattered uniformly over a feasiblesolution space. The IPSO algorithm has been tested on a modeled 10-unit system andthe performance is compared with that of GA and EP methods.Ting et al. [61] have integrated a new approach of hybrid particle swarmoptimization (HPSO) scheme, which is a blend of HPSO, BPSO and real-codedparticle swarm optimization (RCPSO), to solve the UC problem. The UC problem has

een handled by BPSO, whereas the economic load dispatch problem has been solvedby RCPSO.Funabashi et al. [62] have formulated a twofold simulated annealing methodfor the optimization of fuzzy-based UC model. The method has served to offer arobust solution for UC problem.Victoire et al. [63] have applied a hybrid PSO and sequential quadraticprogramming (SQP) technique, prelude to tabu search (TS) method for solving theUC problem. The combinational part of the UC problem has been solved using the TSmethod. The nonlinear optimization part of economic dispatch problem (EDP) hasbeen solved using a hybrid PSO-SQP technique. The effectiveness of hybridoptimization technique has been tested on a NTPS zone-11 7-unit system.1.3.3. REAL POWER LOSS MINIMIZATION AND VOLTAGE STABILITYENHANCEMENTYoshida et al. [64] have applied a PSO technique to reactive poweroptimization problem. The objective has been to find the optimal settings of somecontrol variables that served to minimize the total real power losses in a network. Theproblem has been classified as a mixed-integer nonlinear optimization problem sincesome variables are continuous whereas others are discrete. The test results of anIEEE-14 bus system, a practical 112 bus system and a large scale 1217 bus systemhave been compared with that of reactive tabu search (RTS) and enumerationmethods.Miranda et al. [65] have introduced a hybrid PSO approach to VoltageNArcontrol problem in power systems. They have combined evolutionary strategies withPSO to improve the robustness of the classical PSO. The test results of an IEEE-24bus system have been compared with that of simulated annealing based approach.Mantawy et al. [66] have investigated the same problem using particle swarmoptimization for an IEEE-6 bus system. The simulation results of the PSO algorithmhave been compared with that of previously reported works in the literature.

Ahmed A. Esmin et al. [67] have built a HPSO technique as a modemoptimization tool for real power loss minimization. The technique made use of PSOfor its global search capability to allocate the optimum amount of shunt compensatorsto be installed in each bus. The test results of an IEEE-I 18 bus system have beencompared with that of primal4ual interior point (IP) and genetic algorithms.Tripathy et al. [68] have detailed a novel work on bacteria foraging basedsolution for optimization of real power loss and voltage stability limit. Its mainobjectives are to optimize the transformer taps, UPFC location and its injectionvoltage for a single objective of real power loss minimization, and then for themultiple objectives of loss minimization and voltage stability limit maximization. Thetest results of a 39-bus New England power system have been compared with that ofinterior point successive linearization program (IPSLP) technique.Yair Malachi et al. 1691 have presented a GA-based approach for the selectionof corrective control actions for bus voltage and generator reactive power in a powersystem. The technique has used GA for its heuristic selection of participating controlsof a minimum number of control actions in a distributed load flow environment. Themethod has been successfully applied to a 220-bus model. The test results of GA havebeen compared with that of integer programming based solution method.Amgad A. EL-Dib et al. [70] have proposed a solution technique for findingthe optimum location and sizing of the shunt compensation devices in transmissionsystems with an objective to improve the voltage stability of the system bymaintaining acceptable voltage profile. It has heen solved using newly developedevolutionary-technique PSO. The test results of the Ward-Hale 6 bus, IEEE-14 and -30 bus systems for the proposed method have been compared with that of geneticalgorithm.

13.4. DISTRIBUTED GENERATION @G) FOR POWER QUALITYIMPROVEMENTRamakumar et a]. [71] have proposed an approach to highlight the significanceof voltage profile improvement through the use of various power quality indices. Theproposed indices have served to identify the best location and ratings of DG. It hasalso contributed in the line-loss and environmental impact reductions.Pathomtaht Chiradeja et al. [72] have developed an approach to quantify thetechnical benefits of DG, powered by both conventional and renewable energysources. The benefits have been measured using voltage profile improvement index,line-loss reduction index, environmental impact reduction index and DG benefitindex. The test results of a simple 12 bus system and a radial system have beenillustrated to highlight its usefulness.Joss et al. [73] have demonstrated the potential of DG with power electronicinterface to provide ancillary services such as reactive power, voltage sagcompensation and harmonic filtering. It has proved the ability of DG to compensatevoltage sag resulting from faults in the power system.Chiradeja et al. [74] have evaluated a probabilistic approach based onconvolution technique to quantify the benefit of voltage profile improvementinvolving wind turbine generation.Celli et al. [75] have proposed a GA based software computing procedure toestablish optimum DG allocation on an existing distribution network considering theconstraints such as feeder capacity limits, feeder voltage profile and three-phase shortcircuit. The methodology has been tested on a real MV Italian distribution network.The test results have displayed that considerable savings is achieved by adding someDG units in the appropriate position.Greatbanks et al. [76] have formulated a methodology for locating the mostappropriate site and deciding the size of DGs. The solution algorithm has been testedon several distribution systems representing urban, mal and mixed using 11-kV

networks. The results have demonstrated that it has contributed to impmve bothsystem security and reliability by improving feeder voltage profile, reducing losses,and increasing efficiency.13.5. ENERGY OPTIMIZED DYNAMIC VOLTAGE RESTORERChoi et al. [77] have developed various voltage control strategies for dynamicvoltage restoration with minimum energy injection. It has made use of instantaneousphase advance and progressive phase advance methods for voltage restoration,thereby reducing energy injection. The simulation results of a single phase examplehave revealed the efficacy of the proposed method for balanced and unbalancedvoltage sags.Alexander Domijan et al. 1781 have addressed the potential problems related topower quality devices (PQDs) such as an advanced static VAR compensator, adynamic voltage restorer and a high-speed transfer switch. Its performance has beenmodeled using ATP-EMPTP techniques. Comparison between simulation results andfield measurements of individual PQDs for different system conditions and faultshave been presented.Haque et al. [79] have proposed a method to determine the exact amount ofvoltage injection required to systematically correct a specific voltage drop withminimum active power injection. The technique of correcting the voltage drop or sagby a DVR with minimum active power injection has been tested for lower and highervoltage drops.Mahinda Vilanthgamuna et al. [SO] have built a new phase advancecompensation strategy for DVR in order to enhance the voltage restoration property ofthe device. Supply voltage amplitude and phase detection scheme as well as phaseadvance determination schemes have been included. The efficacy of the proposedmethod has been compared with conventional in-phase injection technique in terms ofenergy saving and dynamic performance.

A new topology based on 2-source inverter for the DVR has been proposed byMahinda Vilanthgamuna et al. [81]. It has served to enhance the capability of theDVR through better utilization of the stored energy. The simulation results haverevealed that the disturbance caused by sag has been effectively compensated utilizingbuck-boost capability of the 2-source inverter.1.4. OBJECTIVES OF THE THESISThe main objectives of the dissertation are:(a) To develop a particle swarm optimization (PSO) algorithm for the economic loaddispatch (ELD) problem in order to minimize generation cost.(b) To generate a PSO-based algorithm for combined economic emission dispatch(CEED) environment for thermal units while satisfying the constraints such asgenerator capacity limits, power balance and line-flow limits.(c) To Integrate PSO method with genetic algorithm for solving ELD problem in apower system having generating units with prohibited operating zones.(d) To solve the unit commitment problem through a PSO-based genetic algorithm.(e) To build a global search technique using PSO with a view to identify optimumsettings of automatic voltage regulator (AVR) values, on-load tap changer(OLTC) positions and the number of reactive power compensation equipments(RPCE) to be connected In order to minimize the real power losses in a powersystem thereby enhancing voltage stability.(f) To maximize the power quality of a power system by proposing a PSO algorithmto obtain the optimum size and location of distributed generations.(g) To propose a novel PSO-based phase advancement compensation strategy forminimizing the energy storage capacity of the capacitor or battery in a dynamicvoltage restorer (DVR) in order to enhance the voltage restoration property of thedevice during voltage sag.

1.5. ORGANISATION OF THE THESISA general introduction to the problem of power system optimization ispresented here in Chapter 1. The need for intelligence based approaches is discussed,and a review of the traditional optimization strategies is traced. It includes a survey ofthe literature and the main objectives of the dissertation.A particle swarm optimization algorithm is applied in Chapter 2 to solve thevarious types of economic load dispatch problems. The proposed work is tested on acombined cycle cogeneration plant system. The results obtained for a 3-, a 6-unitthermal systems and a 3-unit system including a CCCP system are compared withthose obtained using classical and GA-based approaches. The combined economicemission dispatch (CEED) is solved in the same chapter using swarm intelligencebased optimization technique. The efficiency of the PSO algorithm is demonstrated bycomparing the results of an IEEE-30 bus system with that obtained using classical andevolutionary computing techniques. The PSO algorithm is hence applied to a complexCEED problem to obtain minimum generation cost. The chapter also discusses anewly proposed hybrid genetic approach based particle swarm optimization to solvethe economic load dispatch (ELD) problem of a 6-, 15- and 40-unit test cases withPOZ constraints and ramp rate limits.A hybrid genetic algorithm combined with PSO-based approach for obtainingoptimal production cost is suggested in Chapter 3. The algorithm is tested on a 10-unit system.An algorithm that attempts to minimize the real power losses, with a view toimprove the voltage stability of the system is analyzed in Chapter 4. This PSO-basedtechnique uses optimum settings of automatic voltage regulator (AVR). on-load tapchanger (OLTC) values and requires minimum number of reactive powercompensation equipments (RPCE). The voltage stability assessment for IEEE-14, -30, -57, -1 18 bus, Neyveli Thermal Power Station bus and Puducherry bus systems isperformed using a line stability index.

A novel PSO-based approach to quantify the technical benefits of distributedgeneration is proposed in Chapter 5. The algorithm determines the optimal value ofdistributed generation (DG) that can maximize the power quality by reducing the linelosses and increasing the voltage profile of various buses. It is tested on an IEEE-30bus system.A PSO-based phase advancement compensation strategy for a balanced faulttest case is presented in Chapter 6. The strategy aims to optimize the energy storagecapacity of the DVR in order to enhance the voltage restoration property of thedevice. The approach identifies the required value of phase advancement anglecorresponding to minimum power injection fiom the energy storage element such as acapacitor or a battery.The contributions of the dissertation along with the scope for future researchin this area find a place in Chapter 7.

CHAPTER 2PSO BASED ECONOMIC LOAD DISPATCH PROBLEMS2.1. INTRODUCTIONThe main aim of electric power utilities is to provide high-quality. reliablepower supply to the consumers at the lowest possible cost while operating to meet thelimits and constraints imposed on the generating units. This formulates the economicload dispatch (ELD) problem for finding the optimal combination of the output powerof all the online generating units that minimizes the total fuel cost, while satisfying anequality constraint and a set of inequality constraints. As the cost of power generationis exorbitant, an optimum dispatch results in economy.In recent years, with an increasing awareness of the environmental pollutioncaused by thermal power plants, limiting the emission of pollutants is becoming acrucial issue in economic power dispatch. The conventional economic power dispatchcannot meet the environmental protection requirements, since it only considersminimizing the total fuel cost. The multiobjective generation dispatch in electricpower systems treats economic and emission impact as competing objectives, whichrequires some reasonable tradeoff among objectives to reach an optimal solution. Thisformulates the combined economic emission dispatch (CEED) problem with anohjective to dispatch the electric power considering both economic and environmentalconcerns.Practically, the real world input4utput characteristics of the generating unitsare highly nonlinear, nonsmooth and discrete in nature owing to prohibited operatingzones, ramp rate limits and multifuel effects. Thus the resultant ELD is a challengingnonconvex optimization problem, which is difficult lo solve using the traditionalmethods.In this work, particle swarm optimization (PSO) algorithm is proposed tosolve the various t)ipes of economic load dispatch problems in power systems such as

economic load dispatch (ELD) for combined cycle cogeneration plant (CCCP),combined economic emission dispatch (CEED) and the economic load dispatch(ELD) with prohibited operating zones considering ramp rate limits. The feasibility ofthe proposed method is demonstrated on six different systems and the numericalresults were compared with classical and other evolut~onary computing techniques.2.2. ECONOMIC LOAD DISPATCH PROBLEM2.2.1. Problem DescriptionEconomic load dispatch problem is the sub problem of optimal power flow(OPF). The main objective of ELD is to minimize the fuel cost while satisfying theload demand with transmission constraints.2.2.2. Objective FunctionThe classical ELD with power balance and generation limit constraints hasbeen formulated [82] as follows.dMinimize F, = IF, (P,)1-1where Ft is the total fuel cost of generation,F,(P,) is the fuel cost function of ith generator,a,, b,, c, are the cost coefficients of ith generator,P, is the real power generation of ith generator,d represents the number of generators connected in the networkThe minimum value of the above objective function has to be found by satisfytng thefollowing consh.aints.The power balance constraint [82]dCP, =Po +PL (2.3)i-I

where Po is the total load of the system andPL is the transmission losses of the system.The total transmission loss [83]where P,,, and P,,, are the real power injections at mth and nth buses andB, are the B-coefficients of transmission loss formula.The inequality constraint on real power generation P, for each generator [82] isP,"" 5 P, 5 P,"" (2.5)where P,""hd PIm' are, respect~vely, mlnlmum and maxlmum values of realpower allowed at generator I.A. Economic Load Dispatch Problem with CCCPCogeneration units play an increasingly important role in the utility industry.The mutual dependencies of the multiple demand and heat-power capacity of thecogeneration units introduce a complication of integrating the system for economicpower dispatch. The cost characteristics of CCCP system (two 75 MW gas turbinesand one 50 MW steam turbine) [82] is obtained and hence can be found that they arenot differentiable. So the lambda-iterative method will fail in obtaining solution forthe ELD of the above problem. The solution for this problem is obtained byformulating the cost equations by curve fitting technique and implementing theproposed PSO algorithm for the optimal scheduling of generators.

B. Constraint Satisfaction TechniqueTo satisfy the equality constraint of equation (2.3). loading of any one of theunits is selected as the dependent loading Pdu, and its present value is replaced by thevalue calculated according to the following equation [83]:where, Pdu can be calculated directly from the equation (2.6) with the known powerdemand PD and the known values of remaining loading of the generators. Therefore,the dispatch solution always satisfies the power balance constraint provided that Pd.also satisfies the operation limit constraint as given in equation (2.5). An infeasiblesolution is omitted and above procedure is repeated until Pd. lies within its operationallimit. As PL also depends on Pdur an expression for Pl can be substituted in terms ofPI. Pz ,..., Pdu ..... Pd and B, coefficients. After substituting PI in the equation (2.6).the independent and dependent generator terms are separated to obtain a quadraticequation for Pa.. The power balance equality condition is exactly met by solving thequadratic equation for I'd,,.2.2.3. Features of Particle Swarm Optimization (PSO)Particle swarm optimization was first introduced by Kennedy and Eberhart inthe year 1995 [5]. It is an exciting new methodology in evolutionary computation anda population-based optimization tool like GA. PSO is motivated from the simulationof the behaviour of social systems such as fish schooling and birds flocking [84]. ThePSO algorithm requires less computation time and less memory because of itsinherent simplicity. The basic assumption behind the PSO algorithm is that birds findfood by flocking and not individually. This leads to the assumption that information isowned jointly in the flocking. The swarm initially has a population of randomsolutions. Each potential solution, called a particle (agent), is given a random velocityand is flown through the problem space. All the particles have memory and eachparticle keeps track of its previous best position @best) and the corresponding fitnessvalue. The swarm has another value called gbest, which is the best value of all the

pbest. Particle swarm optimization has been found to be extremely effectivein solving a wide range of eng~neering problems and solves them very quickly.In a PSO system, population of part~cles exists in the n-dimensional searchspace. Each particle has certain amount of knowledge and will move about the searchspace on the basis of this knowledge. The particle has some inertia attributed to it andhence will continue to have a component of motion in the direction it is moving. Theknows its location in the search space and will encounter with the bestsolution. The particle will then modify its d~rection such that it has additionalcomponents towards its own best position, pbest and towards the overall best position,gbest. The particle updates its velocity [64] and position [64] using the followingequations (2.7) and (2.8)where V:is the velocity of individual i at iterat~on kk is pointer of iterations,W is the weighing factor,cl, c2 are the acceleration coefficients,Randl( ), Rand2( ) are the random numbers between 0 and 1,S: is the current position of individual i at iteration k,pbest, is the best position of individual igbest is the best position of the groupIn equation (2.7), cl has a range (1.5,2), which is called self-confidence range;cz has a range (2, 2.5), which is called swarm range. The coefficients c, and cz pulleach particle towards pbest and gbest positions. Low values of accelerationcoefficients allow particles to roam far from the target regions, before being tuggedback. On the other hand, high values result in abrupt movement towards or past thetarget regions. Hence, the acceleration coefficients cl and CI are often set to be 2according to past experiences. The term c,Rand, () x (pbest, -Sf) is called particlememory influence or cognition part which represents the private thinking of the

itself and the term c,Rand,( ) x (gbest - S: ) is called swarm influence or thesocial part which represents the collaboration among the particles.In the procedure of the particle swarm paradigm, the value of maximumallowed particle velocity Vm determines the resolution, or fitness, with whichregions are to be searched between the present position and the target position. If Vm"is too high, particles may fly past good solutions. If V""" is too small, particles maynot explore sufficiently beyond local solutions. Thus, the system parameter VmX hasthe beneficial effect of preventing explosion and scales the exploration of the particlesearch. The choice of a value for V*" is often set at 10-20% of the dynamic range ofthe variable for each problem.Suitable selection of inertia weight W provides a balance between global andlocal explorations, thus requiring less iteration on an average to find a sufficientlyoptimal solution. Since W decreases linearly from about 0.9 to 0.4 quite often duringa run, the following weighing function [64] is used in (2.7):where W,is the initial weight,W,, is the final weight,iterm, is the maximum iteration number,iter is the current iteration number.Fig. 2.1 shows a concept of modification of a searching point by PSO [64] andFig. 2.2 shows a searching concept with agents in a solution space. Each particlechanges its current position using the integration of vectors [64] as shown in Fig. 2.2

sk : current searching pointski': modified searching pointV' : current velocityvk": modified velocityV,k,, : velocity based on pbestVgbelt : velocity based on gbestFig. 2.1. Concept of modification of a searching point by PSOFig. 2.2. Searching concept with particles in a solution space by PSOThe equation (2.7) is used to calculate the particle's new velocity according toits previous velocity and the distances of its current posltion from its own bestexperience (position) and the group's best experience. Then the particle flies towardsa new position according to (2.8). The performance of each particle is measuredaccording to a predefined fitness function, which is related to the problem to besolved.The step by step procedure of PSO algorithm is given as follows:I. initialize a population of particles with random values and velocities withinthe d-dimensional search space. Initialize the maximum allowable velocitymagnitude of any particle Vmax. Evaluate the fitness of each particle and assign

the particle's position to pbest position and fitness to pbest fitness. ldentify thebest among the pbest as gbest.2. Change the velocity and position of the particle according to equations (2.7)and (2.8). respectively.3. For each particle, evaluate the fitness, if all decisions variables are within thesearch ranges.4. Compare the particle's fitness evaluation with its previous pbest. If the currentvalue is better than the previous pbest, then set the pbest value equal to thecurrent value and the phest location equal to the current location in the d-dimensional search space.5. Compare the best current fitness evaluation with the population gbest. If thecurrent value is better than the population gbest, then reset the gbest to thecurrent best position and the fitness value to current fitness value.6. Repeat steps 2-5 until a stopping criterion, such as sufficiently good gbestfitness or a maximum number of iterationsifunction evaluations is met.

The general flowchart of PSO is illustrated as follows:Fig. 2.3. Flowchart of PSO methodAdwantages of PSO:I'SO is a population-baaed ecolutionar) technique that has many advantagesover other optimization techniques. PSO is a derivati~e-frec algorithm unlike manyco~nentional techniques and is Iesc smsiti\,c lo the nature of the objective function.v~L.. convexity or continuity. I'hese swarm intelligence hased methods have fewparameters to adjust and escape local minima. I'he proposed method is easy toimplement and program with hasic mathematical and logical operation>. It can alsohandle objective functions with stochastic nature and does not require a good initialst)tution to start its iteration process.2.2.4. lmplementetion of PSO for ELD solutionThe main ohjectibe of ELD is to ohtain the amount of real power to begenerated by each cornmined generator, while achieving a minimum generation Cost

wlthln the constraints. The details of the implementat~on of PSO components aresummarized in the following subsections.2.2.4.1. Representation of an Individual StringFor an efficient evolutionary method, the representat~on of chromosomestrings of the problem parameter set is important [42]. Since the decision variables ofthe ELD problems are real power generations, the generation power output of eachunit IS represented as a gene, and many genes compnse an individual in the swarm.Each individual within the population represents a candidate solution for an ELDproblem. For example, if there are d units that must be operated to provide power toloads, then the ilh individual Pg, can be defined [42] as follows:Pg, =[P,,,P ,. ....... P,,], i=1,2 ..... n (2.10)where n means population size, d is the number of generator, Pbd is thegeneration power output of d'h unit at individual i. The dimension of a population is(n x d). These genes in each individual are represented as real values. The matrixrepresentation of a population is as follows:Individual p, I p12number2.2.4.2. Evaluation FunctionThe evaluation function for evaluating the minimum generation cost of eachindividual in the population is adopted [42] as follows:dMinimizeF, = CF,(P,) (2.11)I-)

2.2.5. Algorithm of the Proposed MethodThe search procedure for calculating the optimal generation quantity of eachunit is summarized as follows:1. In the ELD problems the number of online generating units is the 'dimension' ofthis problem. The particles are randomly generated between the maximum and theminimum operating limits of the generators and represented using equatlon (2.10).2 To each individual of the population calculate the dependent unlt output Pd. fmmthe power balance equation and employ the B-coefficient loss formula to calculatethe transmission loss P1,using constraint satisfact~on techn~que.3. Calculate the evaluation value of each individualp,, in the population uslng theevaluation function f, given by equation (2.1 1).4. Compare each individual's evaluation value with its pbest. The best evaluationvalue among the pbest is identified as gbest.5. Modify the member velocity V of each individual P8, accord~ng to the followingequation:V,:""=WV,"'+c, Rand,( )x( pbest,, -P,,,"')+c,~and,( )x(gbest, - P,~"')i=1,2 ,... n, d=1,2 ..... m (2.12)where n is the population size, m is the generator units.6. Check the velocity constraints of the members of each individual from thefollowing conditions [42]:If v,''"' > Vdmi , then v,,""' = Vdm ,~f v,""' < V," . then v,'"" = V," ,

7. Modify the member position of each individual Pp, [42] accord~ng to the equationp,e''*'l = p; + v:,"') (2.14)P,~'""must satisfy the constraints, namely the generating limits, described byequation (2.5). If P~~'"" violates the constraints, then^^^"'" must be modifiedtowards the nearest margin of the feasible solution.8. If the evaluation value of each individual is better than previous pbest, the currentvalue is set to be pbest. If the best pbest is better than gbest, the best pbest is set tobe gbest.9. If the number of iterations reaches the maximum, then go to step 10. Otherwise, goto step 2.10. The individual that generates the latest gbest is the optimal generation power ofeach unit with the minimum total generation cost.2.2.6. Numerical Examples, Simulation Results and AnalysisThe study has been conducted on test cases with 3-unit thermal, 6-unit thermaland two thermal units with 1-unit as combined cycle cogeneration plant system. Thedescription of the test systems are described in the followmg sections.Test Case 1: Three-Unit Thermal SystemThe cost coefficients of 3-unit thermal system are taken from [82]. The costequations are given below in Rs/h:FI = 0.00156 PI' + 7.92 PI + 561 RdhF2 =0.00194 ~ 2 + ' 7.85 P2 + 310 RdhF3 = 0.00482 P: + 7.97 Pp + 78 Rsh

B, coefficient matrix:The unit operating ranges are100 MW 5 PI 5600 MW ;lOOMW5P25400MW ;50MW5P,5200MW;Test Case 2: Two Thermal Units and One CCCP SystemIn this case, the first two units are the same as 3-unit system and the third unitis replaced with a combined cycle cogeneration plant (CCCP). In CCCP. gas andsteam turbines are working in combination to generate electric power. CCCP has two75 MW gas turbine units and one 50 MW steam turbine unit [82]. The fuel costcharacteristics of this plant is shown in Fig. 2.4Generator Power (MW) --+Fig. 2.4. Fuel cost characteristics of CCCP system

By the method of curve fitting, the cost equation for third plant is formed asfollows.F, =8.517P3 +62.75Rs/h 50 MW 5 P, 5 63.75 MW;= 605.67 Rs/h 63.75 MW 5 P, < 82.875 MW;= 24.08 P, - 1390.04 Rsih 82.875 MW s P, 5 93.75 MW;=9.18P, +6.829&h93.75 MW 5 P, 5 157.5 MW;= 1452.84 Rsih 157.5 MW 5 P, < 176.625 MW;= 17.62P, -1660Rsih 176.625 MW 5 P, 5 200 MW;Test Case 3: Six-Unit Thermal SystemThe system tested consists of six-thermal units [83]. The cost coefficients ofthe system are given below in Rsih:F, = 0.15247~1*+ 38.53973 PI +756.79886 RshFZ = 0.10587~2~+ 46.15916 P2+451.32513 RsihF3 = 0.02803~3~ + 40.39655 P3+1049.9977 RsihF4 = 0.03546~4~+ 38.30553 P4+1243.5311 RsihF5 = 0.0211 1~2+ 36.32782 PS +1658.5596 RshF6 = 0.01799~2+ 38.27041 P6+1356.6592 RshThe unit operating ranges are10 MW 5 PI 5 125 MW; 10 MW 5 P2 5 150 MW;35 MW 5 P3 5225 MW; 35MWSPq5210MW;130MWiP~5325MW; I25 MW 5Pg 5315 MW;

B,. Coefficient matrix:To verify the feasibility of the proposed PSO method, three different powersystems were tested, under the same evaluation function and individual definition. 50trials were performed to observe the evolutionary process and to compare theirsolution quality, convergence characteristic, and computation efliciency. From theexperiences of many experiments the following parameters are selected for theparticle swarm optimization algorithm to solve the above test cases and are given inTable 2.1. For implementing the above algorithm, the simulation studies were carriedout on P-IV, 2.4 GHz, 512 MBDDR RAM system in MATLAB environment.Table 2.1. Parameters used in PSO method - 3.6 and CCCP unit systemsParameterValuePopulation size10Wms0.9Acceleration Coefficients cl,cz2.0.2.02.2.6.1. Test Case 1: Three-Unit Thermal SystemThe economic load dispatch for the first test case with the corresponding loadsis given as 585 MW. 700 MW znd 800 MW, respectively. The proposed PSO methodis applied to obtain the minimum generation cost. Table 2.2. provides the results ofoptimal scheduling of generators obtained by proposed PSO method for three thermalunits system with losses neglected. Table 2.3. provides a comparison of economicload dispatch results obtained by various optimization methods for a three unitthermal system with losses neglected.

Table 2.2. Optimal scbeduliig of generators of 3-unit system neglecting lossesTable 2.3. Solution of different methods neglecting losses - 3-unit systemSl. No.1.Load DemandPo(MW)585Conventional MethodIS21(bh)5821.4000GAMethodls2'(Rsh)5827.5PSOMethod(Rsh)5821.42.7006838.40566877.26838.43.8007738.5 1897756.87738.5The above system is solved for a load demand of 585.33 MW using theproposed PSO method with the inclusion of transmission loss. The optimal schedulingof generators obtained by the PSO algorithm for a three-unit thermal system wasshoun in Table 2.4. By following the above procedure, the solution obtained by theproposed method for a three unit thermal unit system with losses included is given inTable 2.5.

Table 2.4. Optimal scheduliig of generators including losses - 3-unit systemr,Table 2.5. Solution of different methods including losses - 3-unit systemLoadConventional Method GA Method Proposed1821 1821 PSO Method(&h) (M)Deyd (MW)Fig. 2.5, shows the convergence characteristics of the proposed algorithm for aload demand (PD) of 585 MW with losses neglected. Fig. 2.6. shows the reliability ofthe proposed algorithm for different runs of the program. The figure shows that thealgorithm is capable of obtaining a faster convergence for the three unit thermalsystem in a very few generations and the solution is consistent.5865Convergenece PropertyLsr,--d=.51190.SIS.1105 LO IS 20 l5Number of GenerationsFig. 2.5. PSO based ELD convergence characteristics -3-unit system

5-5 1 0 1 5 m U m 3 5 . 0 U 5 0Nmnber of RunsFig. 2.6. Reliability characteristics evaluation - 3-Unit System2.2.6.2. Test Case 2: Three-Unit System with CCCPThe economic load dispatch is solved using a proposed PSO algorithm for athree unit system with CCCP having system load as 680 MW, 750 MW and 869 MW,respectively. Table 2.6. summarizes the optimal dispatch of load among the availablegenerating units. The simulation results were studied and the obtained values of costof generation of different methods are given in Table 2.7. The cost was found to beminimum in the PSO based method.Table 2.6. Optimal scheduling of generators including CCCP - 3-unit system

Table 2.7. Solution of diierent metbods including CCCP - 3-unit systemSI. No.Load Demand(MW)GA Method(Rs/b)1821ProposedPSO Method2.2.6.3. Test Case 3: Six-Unit Thermal SystemThe third case of a six unit thermal system is solved by the PSO method andthe optimal scheduling of generators for the load demands of 700 MW and 800 MWis tabulated in Table 2.8. The results obtained by the proposed algorithm arecompared with other evolutionary computing techniques and are given in Table2.9.Table 2.8. Optimal scheduling of generators - 6-unit systemsl.Vc)laadI>cmand(MW)P,(MW)P,(MW)P,(MW)P,(MW)PI(MW)P,(MW)I;,(Rslh)h a sPL(MW)Excrulic~nl'lmr (Srr)1.70016.5624.73138.2116.6208.4214.23698718.81.1722.800251211618228720342114261.609

Table 2.9. Solution odiffereut methods - Cuoit systemFuel Cost 1LossPI. ..Execution Time1 tconventional method~331IReal-coded GAh d37l37:37288.70123.13426.570.2514.611 proposed PSO method 1 36987 1 18.84471 1.17Hybrid GA method[83137137.9623.124 1.21From the comparison of results for the test cases, it is demonstrated that theproposed algorithm performs better than the conventional, genetic algorithm andrcfined genetic based algorithm methods in the aspect of reduction of fuel cost as wellas real power loss.2.3. COMBINED ECONOMIC EMISSION DISPATCH PROBLEM2.3.1. Problem DescriptionThe optimum economic dispatch may not be the best in terms of theenvironmental criteria. Harmful ecological effects by the emission of gaseouspollutants from fossil fuel power plants can be reduced by proper load allocationamong the various generating units of the plants. But this load allocation may lead toincrease in the operating cost of the generating units. So, it is necessary to find out asolution which gives a balanced result between emission and cost. This is achieved bycombined economic emission dispatch problem. This dual-objective CEED problemis converted into a single objective function using a price penalty factor approach[461.

2.3.2. Objective FunctionOptimization of generat~on cost has been fomiulatrd based on classical E1.Dwith emission and line flow constraints. The deta~led problem is gluen 1161 asfollows.where F is the optimal cost ofgenerat~on.FC and EC are total fuel cost and emission costs of generators, respect~vely.d represents the number of generators connected in thc networkThe minimum value of the above otyect~ve funct~on has to bc found oursubject to constraints given by Eqs (2.3) and (2.5)The power flow equatlon of the powcr network (461where P, and Q, are calculated real and reactive power for PQ bus i.respectively;P,"" and Q,"" are specified real and react~ve power for PQ bus i,respectively;P,.I,, and Pz,, arc calculated and specified real power for PV bus m,respectively;I v ~ and 6 are voltage magnitudes and phase angles of different buses.The inequality constraint on voltage of each PQ bus [46]V- (I) S V, S V- (I)

where V, (i) and VmX(1) are minimum and maxlmum voltage at bus I.respective1 yThe maximum power limit on transmrssion line 1461 IS given bywherenl represents number of lines,~f d"MVA 1s the calculated l~ne flow of each transmlsslon line.~f,,"~""* is the rated llnc flow of each transmission line.Total fuel cost of generation FC In terms of control variables generator powcrscan be expressed [46] as follows.where P, is the real power output of an iU' generator inMWi represents the corresponding generator.a,, b, ,c, are the fuel cost coefficients of generators.The total emission release can be expressed [46] aswhere a, ,p,,y, are emission coefficients of generatorsThe dual-objective combined monomic emission dispatch problem isconverted into single optimization problem by introducing a price penalty factor h[92] as follows.Minimize 0, = FC + h x EC $/h (2.22)

subjectad to the power flow constraints of equation (2.3, 2.5, 2.17. 2.18 and 2.19).The price penalty factor h blends the emission with fuel cog and a, is the totaloperating wst in $h.The price penalty factor h, is the ratio between the maximum fuel cost andmaximum emission of corresponding generator.The following steps are used to find the price penalty factor for a particularload demand:1. Find the ratio between maximum fuel cost and maximum emission of eachgenerator2. Arrange the values of price penalty factor in ascending order.3. Add the maximum capacity of each unit (P,'"lu) one at a time. starting from thesmallest h, until Po"" > P,, .4. At this stage, h, associated with the last unit in the process is the price penaltyfactor h for the given load.The above procedure gives the approximate value of price penalty factorcomputation for the corresponding load demand. Hence a modified price penaltyfactor (h,) is used to give the exact value for the particular load demand. The first twosteps of h computation remain the same for the calculation of modified price penaltyfactor. Then it is calculated by interpolating the values of h, corresponding to theirload demand values.2.3.3. Step-by-step algorithmThe step-by-step algorithm for the proposed method is explained as follows:1. Specify the maximum and minimum limits of generation power of eachgenerating unit, maximum number of iterations to be performed and fuel costcoefficient of each unit.

2. Initialize randomly the individuals of the population according to the limit ofeach unit including individual dimensions, searching points, and velocities.These initial individuals must be feasible candidate solutions that satisfy thepractical operation constraints.3 To each chromosome of the population the dependent unit output Pd will becalculated from the power balance equation and B-coeficient loss formula isemployed to calculate the transmission loss Pi.4. Calculate the evaluation value of each population P,, using the evaluationequations (2.20) and (2.21).5. Calculate the price penalty factor using the equation (2.24).6. Compute the new evaluation function using the equation (2.22).7. Compare each population's evaluation value with its pbest. The bestevaluation value among the pbest is denoted as gbest.8. Modify the member velocity V of each individual P,, according to thcequation (2.1 2).9. Check the velocity components constraint occurring in the limits using theconditions (2.1 3).10. Modify the member position of each Individual P,, according to the equation(2.14). If P,,d'"'l must satisfy the constraints, namely the generating limits.described by (2.5). If Pgdl''" violates the constraints, then P,~"'" must bemodified towards the near margin ofthe feasible solution.I I. If the evaluation value of each population is better than the previous pbest thecurrent value is set to be pbest. If the best pbest is better than the gbest thevalue is set to be gbest.12. If the number of iterations reaches the maximum then go to step 13, otherwisego to step 3.13. The individ-d that generates the latest gbest is the optimal generation powerof each unit.14. After obtaining the global optimum solution, power flow is computed usingNewton-Raphson method and the calculated MVA of line flow is comparedwith the rated MVA of line flow.

IS. If the line is found to be overloaded previous gbcst value is chosen as theglobal optimum solution.16. Stop2.3.4. Simulation ResultsThe proposed algorithm is applied to an IEEE-30 bus system. The total systemload demand is 283.4 MW whose data has been given in Appendix-A. In the proposedapproach minimum generation cost of the generating units was obtained using PSObased ELD in CEED environment. The line flows in the system was compared usingNewton-Raphson method. The simulation studies were carried out using a P-IV 2.4(;Hz. 512 MB DDR RAM system in MATL.AB environment. Tahle 2.10. provides thesimulation parameters of the proposed PSO algorithm. The execution time for thePSO based method is obtained as 79.9220 seconds. Fig. 2.7. shows the convergencecharacteristics of PSO based ELD algorithm. The maximum. average and minimumcost of generation are presented for an IEEE-30 bus system. Tahle 2.1 1. summarizesthe minimum solution obtained by particle swarm optimization based economic loaddispatch with line flow constraints for the bus system. The minimum solution includesoptimum generations, total loss. total fuel cost for IEEE-30 bus system. Table 2.12.shows the comparison of optimal generation schedule obtained by the PSO basedmethod with other evolutionary techniques. The best generation of IEEE-30 busgenerating units obtained from PSO-CEED algorithm and is presented in the Fig. 2.8.Table 2.10. Parameters used in PSO method - IEEE-30 bus system7 ParameterPopulation sizeINumber of iterationsWmWmmAcceleration coefficients, c~,czValue501000.90.42.0,2.0

Table 2.1 1. ELD results obtained h? various methods - IEEE-30 bus systemMethod1.u-1~~ 1x51O\erall cost (SIh)805 4500Loss (MW)I I o~oo1rphlc 2.12. Minimum power dispatch results h? \nriou\ methods - IF:EI

Gansrnror N-bsrFig. 2.8. Best generator settings of PSO-based

The computational procedure of price penalty factor for IEEE-30 bus systemis explained as follows. The ratio between the maximum fuel cost and minimumemission of six generating units are found and arranged in ascending order.h,= [h, hq h6b3 h~ hllh, = [I .7707 1.7916 2.0534 2.2198 2.3310 2.34361

The corresponding maximum limits of generating units are given byp,- = [30 35 40 50 80 2001For a load of Po MW starting from the lowest h, value, the maximum capacityof the units is added one by one (m) and when this total equals or exceeds the load. h,associated with the last unit in the process is price penalty factor [92].rn = [30 65 105 155 235 4351For PD = 283.4 MW, (30+65+105+155+235+435) MW>283.4MW.Hence price penalty factor (h) is determined as 2.3436 for IEEE-30 bussystem. Even though the price penalty factor was computed for 283.4 MW but it givesthe value up to a load demand of435 MW. So the modified price penalty factor 1461IS computed by interpolating the values of h, for the last two units by satisfying thecorresponding load demand.where h, is the modified price penalty factor in $/kg,h,, is the price penalty factor associated with the last unit in $/kg.h,2 is the price penalty factor associated with the current unit in $/kg.P,,, is the Maximum power associated with the last unit in MW,PmaS is the Maximum power associated with the current unit in MWBy following the above procedure, the minimum solution was obtained by theproposed PSO method for IEEE-30 bus system as given in Table 2.14. Byincorporating modified price penalty factor approach, the total operating cost of1624 $/h was obtained. The convergence characteristics for the PSO method areshown in Fig. 2.9.

Table 2.14. CEED raub - IEEE-30 bus systemPriceEmissionPenaltyFactorFCPSO2.3340835.5655 377.2407 1624 5.664Number orCanadonrFig. 2.9. PSO-based CEED convergence characteristics -IEEE-30 bus system2.1. ECONOMIC DISPATCH PROBLEM WITH PROHIBITEDOPERATING ZONES2.4.1. Problem DescriptionEconomic dispatch is an important daily optimization task in the power systemoperation. Large modem generating units with multivalve steam turbines exhibit alarge variation in the input-output characteristic functions. Thus, the practical EDplanning must perform optimal generation dispatch among the generating units to

sat~sfy the system load demand and practical operation constraints of generators thatlnclude the ramp rate limits and the proh~bited operating zones.2.4.2. Objective FunctionThe objective of ED problem IS to minlmlze the total fuel cost of power plantssubjected to the operating constraints of a power system. Generally. ~t can heformulated with an objective function. subject to the practical operation constraints ofgenerator [42].dMinimize F, = F, (P, ) (2.25)1-1where F, is the total generation cost,F,(P,) is the cost function of I ' ~ generator.a,,b,,c, are thc cost coefficients of iIh generator.P, is the real power generation of iIh generator,d represents the number of generators connected in the network.The minimum value of the above objective function has to be found out bysat~sfying the following constraints.Power balance Constra~nt, in whichThe cost is optimized with the following power system constraint [42]where PD is the total load of the system andPL is the transmission power loss of the systemRamp rate limit constraintIn ELD research, a number of studies have focused upon the economicalaspeas of the problem under the assumption that unit generation output can be

adjusted instantaneously. Even though this assumpt~on s~mplifies the problem. it doesnot reflect the actual operating processes of the generating un~t. The operating rangeof all on-line units is restricted by their ramp rate l~mits [SO]. Fig.2.10. shows threesituations when a unit is on-line from hour 1-1 to hour I. Fig 2.1qa) showsthat the unit is in a steady operating status. F I 2.1qb) ~ shows that the unit is in an~ncreasing power generation status. Fig. 2.1qc) shows that the unit is in a decreasingpower generation status.P' 'I' --(a) @) (c)Fig. 2.10. Three possible operating conditions of a generating unit(i) as generation increases [42]P, -pan S UR,(ii) as generation decreases [42]P," - P, < DR,where P, is the power generation of unit I,P,' is the power generation ofunit i at previous hour,UR, is the ramp rate limit of unit i as power generation increases andDR, is the ramp rate limit of unit i as power generation decreasesThe ramp rate constraints restrict the operating range of the physical lower andupper limit to the effective lower limit P,m' and effectwe upperrespectively [42].limit^,'"",Hence the ramp rate constraint is stated [42] asProhibited Opaating Zone Constraint, in which references [42] have shownthe input-output performance curve for a typical thermal unit with many valve

points. These valve points generate many prohtblted zones. For a unit withprohibited operating zones, the zones dlvlde the opcratlng region between theminimum generation limit (P"") and the maximum generation limil (PMU).These prohibited operating zones are due to phystcal limitations of powerplant components such as vibrations In a shatt heanng which amplified in acertain operating region. For a prohlblted zone. the unit can only operateabove or below the zone [42].where z is the number of prohibited zones of a unlt,P,L,P,'> are the Lower and Upper llmlts of z proh~h~ted zones of unit i(MW).11 is the lower limlt of the prohibited operating zone of unit i,ul is the upper limit of the prohibited operating zone of untt i,d is the number of generating units,P,"" is the effective lower limit of iCh unit with ramp rate constraint,P,"" is the effective upper limit of iIh unit with ramp rate constraint.The total transmission network losses is a function of unit power outputs thatcan be represented using B coefficients [42]:where P, and PJ are the real power injections at i* and j" buses,respective1 y,B, are the B-coefficients of transmission loss formula,B, is the vector of same length as generators,Boo is a constant.

2.43. Futures of Genetic AlgorithmsA global optimization technique known as genetic algorithm (GA). aprobabilistic and heuristic approach is used to solve power system optimizationproblems. Genetic algorithm. unlike strict mathematical methods. has the apparentability to adapt to nonlinearities and discontinuities commonly found in theoptimization problems. Genetic algorithms are attractive and serve as an alternativetool for solving combinational optimization problems and they are found to hcsuperior in their parallel search ability that climbs many peaks in parallel.Genetic algorithms are adaptive heuristic search algorithms premised on theevolutionary ideas of natural selection and genetics (41. The basic concepts of GASare designed to simulate processes in natural system. necessary for evolution.specifically those that follow the principles first laid down by Charles Darwin ofsurvival of the fittest. As such they represent an intelligent exploitation of a randomsearch within a defined search space to solve a problem.First pioneered by John Holland in 1960. Genetic algorithms have been widelystudied. experimented and applied in many fields in engineering world. Not only doesCiAs provide alternate methods for solving problems but it consistently outperformsother traditional methods in most of the problem links. Many of the real worldproblems involved in finding optimal parameters. which might prove difficult fortraditional methods are ideal for GA's 1491. They can solve problems that do not havea precisely defined solving methods. or if they do. when following the exact solvingmethod would take far too much time. The genetic algorithm on the other hand, worksonly with the objective function information in a search space for an optimalparameter set.2.43.1 Components of Genetic AlgorithmsGAS are derived from a simple model of population genetics. They have thefollowing five components (861:(i)String representation of the control variables

(ii) An initial population of strings.(iii) An evaluation function that plays the role of the envimnment. rating thestrings in terms of their fitness i.e., their ability IO survive.(iv) Genetic operators determine the composition of a new populationgenerated from the previous one by reproduction. crossover and mutation.(v) Value of the parameters that the GAS use.Since GAS are based on nahd genetics, there exists strong analogies betweengenetic algorithm and natural genetics. The strings are similar to chromosomes inbiological systems, where the chromosomes are composed of genes, which may takeany of several forms called "alleles" 1871. If the control variables are represented inbinary bits and concatenated to form a string, then it is called as binary coded GA andif the control variables are represented in real numbers. then it is called as real codedGA. GAS do not work with a single string but with a population of strings. whichevolves iteratively by generating new strings taking the place of their parents. GAStreat the problem, as a black box in which the input is the strings and the output istheir fitness.Three basic operators comprise a GA. They are reproduction, crossover andmutation. Reproduction is the mechanism by which the most highly fit members in apopulation are selected to pass on information lo the next population of members. Iteffectively selects the finest of the strings in the current population to be used ingenerating the next population. In this way. relevant information concerning thefitness of a string is passed along to successive generations. It can be shown that GASactually allocate exponentially increasing trials to the most fit of these strings.Crossover serves as a mechanism by which strings can exchange information.possibly creating more highly fit swings in the process and allowing the exploration ofnew regions of the search space 1871. Many types of crossovers are available likesingle point crossover, multipoint crossover, uniform crossover and windowcrossover. The last of the GA operators is mutation, and is generally considered as asecondary operator. Mutation ensures that a string position will never be fixed at acertain value for all time. Like other stochastic methods, GAS require a number ofparameters, which are population size, probability of crossover, probability ofmutation. Usually small population size, high crossover probability and low mutationprobability are recommended [88].

Tbe GA's can be distinguished hrn other optimization methods in fourdifferent ways as follows:(i) GA's use objective function information to guide the search, not thederivatives or other auxiliary information.(ii) GA's use a coding of the parameters used to calculate the objectivefunction in guiding the search, not the parameter themselves.(iii) GA's search through many points in the solution space at one time. not asingle point.(iv) GA's use probabilistic rules, not deterministic rules, in moving from oneset of solutions (a population) to the next.The sofi computing techniques for optimization are mainly hased on GA.Though the GA methods have been employed successfully to solve complexoptimization prohlems, recent research has identified some deficiencies in GAperformance. This degradation in efficiency is apparent in applications with highlyep~.s/a/ic objective functions (i.~.. whcre the parameters being optimiied are highlycorrelated) [the crossover and mutation operations cannot ensure betier fitness ofoffspring because chromosomes in the population have similar suuctures and theiraverage fitness is high towards the end of the evolutionary process] 1891. 1901.Moreover. the premature convergence of CiA degrades its performance and reduces itsaearch capability that leads to a higher prohahility toward obtaining a local optimum1x91.2.4.4. CA and PSO Combined Hybrid Method2.4.4.1. Deseription of the proposed methodIn this work, a hybrid genetic-PSO search (hybrid GA-PSO) algorithm isproposed dich utilizes Genetic algorithm to explore the high performance region insolution space and PSO algorithm to exploit the solution space for locating theoptimal solution. Thus, the GA guides PSO for better performance in the complexsolution space. In this work, the constrained economic dispatch problem is solved bythe integrated GA-PSO algorithm and a high quality solution is obtained for apractical power system opemtion. The GA-PSO algorithm is utilized to determine the

optimal generation power of each unit that was usad in operation at the spacificperiod, thus minimizing the total generation wst.2.4.4.2. Evaluation FunctionThe evaluation function f (it is called fitness in GA) must be defined forevaluating the fitness of each individual in the population. For emphasizing the bestchromosome and faster convergence of the iteration process, the evaluation value isnormalized in the range between 0 and 1. The evaluation function f is given inequation (2.34) which is the reciprocal of the summation of generation cost functionF,,,, and power balance constraint P*L 1421.1f=-F,,,, + p,,whereFC", = I + abs(F~." - F,"," )F,,and F,,,, are the maximum and minimum generation cost among he~ndividuals in the initial population.2.4.4.3. Application of GA-PSO AlgorithmThe sequential steps of the proposed hybrid GA-PSO algorithm are shown asbelow:1. Initialize randomly the individuals of the population according to the limit ofeach unit including individual dimensions, searching points, and velocities.These initial individuals must be feasible candidate solutions that satisfy theoperation constraints.

2. To each chromosome of the population the dependent unit output Pd will becalculated fmm the power balance equation and B coefficient matrix.3. Calculate the evaluation value of each individual P, in the population usingthe evaluation function f given by equation (2.34).4. Compare each individual's evaluation value with its pbest. The best evaluationvalue among the pbest's is denoted as gbest.5. Modify the member velocity V of each individual PK according to the equation(2.12).6. Check the velocity components constraint occurring in the limits using theconditions (2.13).7. Modify the member position of each individual P,, using (2.14). if P,~"*"violates the constraints, then \d"*"must be modified toward the near marginof the feasible solution.8. Apply the genetic operators selection. crossover and mutation to the abovepopulation and generate offspring. Now compare the parents and offspring toselect the fittest chromosomes for the next step.9. If the evaluation value of each individual is better than previous pbest. thecurrent value is set to be pbest. If the best pbest is better than gbest. the valueis set to be gbest.10. If the number of iterations reaches the maximum. then go to step 11.Otherwise. go to step 2.11. The individual that generates the latest gbest is the optimal generation powerof each unit with the minimum total generation cost.2.4.5. Simulation ResultsTo verify the feasibility of the proposed hybrid algorithm, the 6-, 15- and 40-unit systems are tested. The ramp rate limits and prohibited operating zones of theunits were taken into consideration. The proposed hybrid GA-PSO algorithm iscompared with the GA and the PSO methods. Under each sample system, 50 trialswere performed using the same evaluation function with the proposed algorithm. Thisgives a fair comparison of the proposed GA-PSO method with the aspects ofComputational efficiency and a qualitative solution. An optimal range of inertia

weight and acceleration factors for the PSO algorithm is estimated for 6, 15- and 40-unit test cases in this research. On comparison of the results. it has been demonstratedthat the proposed algorithm is capable of obtaining higher quality of solutionefficiently for economic dispatch problem covering prohibited operating zones.AAer many experiments. the following parameters have been selected for theproposed hybrid GA-PSO algorithm.I-.Table 2.15. Parameters used in GA-PSO method - Cunit, 15-unit, 40-unitParameterGenerationsWmmWmsnsystemsAcceleration coefticients ~ 1 . ~ 2Crossover probabilityMutation probabilityValuePo~ulation size I 501000.90.42.0.2.00.550.012.4.5.1. Six-Unit SystemThe system contains six thermal units. 26 buses and 46 transmission lines andthe data were taken from [42]. The cost coeficients, generating unit capacity limits.ramp rate. prohibited operating zones and loss coeflicients are provided inAppendix-B.The load demand is 1263 MW. To simulate this system, each individual P, containssix generator power outputs. Since one unit is considered as a dependent unit. eachindividual in the population contains five generator power outputs. The dimension ofthe population is equal to 50x5. Table 2.16 shows the best solution obtained by theproposed algorithm and its comparison with the GA and PSO methods. Table 2.17shows the comparison of average generation cost and average CPU time of differentmethods.

Table 2.16. Optimal generator dispatch solution by various methods - bunitsystemHybridGA Method PSO MethodOutput Power (MW)CA-PSO(421 1421MethodTable 2.17. Comparison of solution quality - 6-unit systemMethodGA method [42]PSO method[421Hybrid GA-PSO1 method---Minimum15,45915,45015.444Generation Cost (Sh)Maximum15,52415.49215,488-Average15.46915.45415.45 1AverageCPll Time(see)41.5814.8912.72The comparison of the performance of proposed algorithm with other methodsdemonstrates that the solution quality and computation efficiency is good for thehybrid algorithm.

2.4.5.2. Fifteen-Unit SystemThis system contains 15 thermal generating units whose characteristics andtransmission loss (B loss) coefficients are Inken from [42] and are provided inAppendix

MethodGA method[421PSO method1421Hybnd GA-PSOmethodTable 2.19. Comparison of solution quality - 1Sunit systemMinimum33,11332,85832.724---Generation Cost (yb)- -..Maximum Average33.337 13.228- -33,331 73,039.- - -- -. --77,188 72,984-Averageexecutiontime (Sec)49 3126 5927 52--For a power demand of 2630 MW, the total transmlsslon losses are 31.75 MWthe optimal dispatch is obtained by the proposed algorithm. Since the total poweroutput from all the 15 units are 2661.75 MW. the power balance equatron is exactlysat~sfied. The total generation cost as well the power losses are less for Ihe proposedalgorithm compared with GA and PSO methods By comparing the results obtainedhy various methods, it is found that the proposed algorithm is capable of providingoptimal solution.-2.4.5.3. Forty-Unit SystemThe system consists of 40 units ~n the realistic Taipower system which is alarge-scale and mixed-generating system with coal-fired, oil-fired, gas-fired, dieseland combined cycle cogeneration units [50]. The cost coeficients of Taipower 40-unit are shown in Appendix-D. The system load demand is 8550 MW. Since one unit1s considered as a dependent unit, each individual in the population contains 39generating unlt outputs. The dimension of the population is 50 x 39. The simulationresults and a comparison of performance are given in Tables 2.20 and 2.21.

Table 2.20. Tat results of the proposed approacb- 40-unit system- -J I.ICA Method I M , r o d 1Output Power (MW)1 ,421Total power output (MW)Power loss (MW)8641.0889.768637.2687.248636.58286.58Total generation cost ($/h)135,070130.380130,255Table 2.21. Comparison of solution quality - 40-unit system1 1421MethodGA methodGeneration Cost (Sth)Minimum Maximum Average135.070 1 137,980 1 137.760 181.80AverageCPU Time(a=)I GA-PSO 130,255( method 1 1 I I ISolution QualityThe developed software package has ken executed 50 different runs Ibr theproposed hybrid GA-PSO algorithm and the results are given in Tables 2.16 to 2.20.From the comparison of results shown in the tables. it is evident that the proposedhybrid GA-PSO algorithm for three systems has obtained low generation cost incomparison to the PSO and GA methods. Fig. 2.1 1 shows the plot of convergence ofbest solution obtained by the proposed algorithm for a 15-unit system. There is anindication of a better quality of solution obtained by the proposed method whencompared to other methods.

Fig. 2.1 1 (;A-IDSO contergence charncteristicc - 15-unit r?rteml.rom the cornp;lrlson 01' l ahler 2 17. 2 1') and 2.21. II can he li>und th;~t thep~opo\ed h>brid (;.A-I30 ha\ Icswr a\eragc e\ccution time compred with (;A illid1'50 nictliods. I lence. the cc~niputation ~.flicicno of' the proposed algorithm I\ruccesslull! dcmonbtrated. Prom the ah~be \tud! ~t IS fi~und that. even thoughc\ccution time li)r one iteration ia more hecaube of the presence of CiA operators.crosso\.er and mutation. the identilication of'the high perl'bnnance region and locatingthe optimal solution is pc>ssihle mithin less numher 01. ~terations tlerse it finds theoplirnal solution within lesser average euccutlcm time.

2.5. CONCLUSIONThis work adapts the particle swarm optimization algorithm and geneticapproach based particle swarm optimization algorithm to different types of economicdispatch problems. The test results for the lEEE and the various test systems bring outthe advantage of the proposed method. The convergence abilities of the PSO methodare better than the classical evolutionary programming method. The PSO methodconverges to the global or near-global pint, irrespective of the shape of the costfunction. for example, discontinuities in the cost functions. The better computationefficiency and convergence property of the proposed PSO approach shows that it canhe applied to a wide range of optimization problems.

CHAPTER 3INTEGRATED GA-PSO BASED UNIT COMMITMENT3.1. INTRODUCTIONUnit commitment plays an important role in the economic operation of apower system. The generator scheduling problem involves the determination of thestart up 1 shut down times and the power output levels OF all the generating units atcach time step over a specified scheduling period T. so that the total stRI-c up, shutdown and running costs are minimized. The standard unit commitment problem issubjected to several constraints that include minimum uptime and down-time, crewconstraints. ramp rate limits, generation constraints. load balances. must-run units andspinning reserve constraints 193-941.A number of conventional methods such as priority list method 1481. linearprogramming method 1951, branch and bound technique [48], lagrangian relaxation1971, dynamic programming [95] and simulated annealing 1981 methods have beenproposed previously for solving unit commitment. 'The ideal method of solving theunit commitment problem involves an exhaustive study of all the possiblecombinations of units to meet the load demand and the combinations corresponding tominimum generation cost and then choosing the best. This straightfoward methodwould test all combinations of units that can supply the load and reserve requirements.The combination that has the least operating cost is taken as the optimal schedule.Given enough time, this enumerative process is guaranteed to find the optimalsolution but the solution must be obtained within a stipulated time that makes it usefulfor the intended purpose. When the problem is highly constrained, the efficiency ofthe solution is poor except for the simplest cases.Extensive works have been done in the unit commitment field for the past twodecades. N.P. Padhy has summarized the previous researches for solving the UCproblem [99]. All the conventional methods only provide near global optimal

solutions and the quality of each solution is affected by either the solution timelimitation, or the feasibility of the final solution. Computer storage requirement is themajor limiting factor but this is fast becoming a thing of the past as the cost ofcomputer memory continues to drop. Recently, there is an upsurge in the use ofmethods such as genetic algorithms. simulated annealing. evolutionary programmingand particle swann optimization that mimic natural processes to solve complexoptimization problems. The above said methods are very efficient for solving highlynonlinear and combinatorial optimization pmblems. When the size of the problem~ncreases, these evolutionary methods will locate the high performance region of thesolution space at quick execution time but they face difficulty in locating the exactoptimal solution.In this research work, a hybrid algorithm which integrates genetic algorithmand PSO method has heen proposed for the solution of unit commitment problem. Thegenetic algorithm is applied to find the ON-OFF status of all the avnilable generatingunits while PSO method solves the economic load dispatch among the committedunits in each and every hour.This chapter presents the application of the prnposcd Integrated Genetic-PSOhascd unit commitment algorithm fc)r a 10-unit system and the results are presented.I'he production cost obtained using proposed GA-PSO method is considerably lesscompared to that of genetic algorithm and BPS0 based unit commitment problems. Acomprehensive unit commitment software package is developed using MATLAR.3.2. PROBLEM DESCRIPTIONUnit Commitment is a combinatorial optimization prohlem which is used inpower systems to properly schedule the generating units such that both the forecastedload demand and operational constraints of the system are satisfied at minimumoperating cost. The operating cost of thennal power system wnsists of fuelexpenditure for operating a unit startup and shutdown costs. The fuel wst of thermalunits exhibits quadratic characteristics. Thermal units may he started up from one ofthe two possible states namely wld start or hot start. Hence, stari up cost depends

upon any one of the two possible states. The operating cost also depends upon thevanous operat~onal constraints of the thermal power system wh~ch makes ~t nonlinear.3.2.1. Objective FunctionThe objective of unlt commitment is to devclop the most cconom~cal stari uparid shut down schedule for all the available generating unlts In the power system thatsatisfies the forecasted load demand and the unit's operating requirements over thescheduling period [loo]. The objective function of the thermal UC problem iscomposed of the fuel and starl-up cost for the generating unlts and can bc expressed[59] asdNhPC = 11 [I, (t) F, ( P, (0) + S, (t)] (3.1)1.1 ,-I'Thc cost of each unit is given [59] as followswhere 1, (1) is the commitment status of ith generator at hour 1,F, (P,(1)) is the fuel cost of ilh generator at hour t ($/h),P, (1) is the power generated in ith unit at eh hour (MW).S, (1) is the start-up cost of iIh unit at hour t ($/h),d is the number of generating units,Nh is the total schedule time (24 h),c,, b,, a, are the cost coefficients of unit i ($/h, S/MWh. SIMW~')Depending on the nature of the power system under study, the unitcommitment problem is subjected to many constraints [101]. The main wnstraintsinclude load balance, generation limits, thermal, time, fuel and economic constraints.

Subject to the following constraints(i) Power demand constraint in order to satisfy the forecasted load demand.the sum of all the generating units on line must be equal to the system loadover the time concerned. It is given [59] by,where Pu(t) is the load demand at hour t (MW).P,(t) is the real power produced by the ith generator athour't'(MW)(~i) Capacity limit constraints in which the real power generated hy the unitmust bc within the maximum and minimum generation limlt of the unlt forachieving economic operation 1591where P,'"" and P,"" are mlnimum and maxlmum power l~m~ts of I"'unit, t=1,2 ....... Nh(iii)Minimum Up Time (MUT) in which once the unit is started up. it shouldnot be shut down before a minimum up-time [59]where T., , is the minimum uptime of i~ generator (h),T,. , is the time duration for which ith generator is continously ON01)-(iv)Minimum Down Time (MDT) in which once the unit is shutdown, itshouldnot be Medup before a minimum down-time [59]

where Tdom, is the minimum down time of generator i (h),Tor., is the duration for which i" generator is continously OFF01).(v) Cold Start HoursIn thermal units the boiler gets cooled to normal temperature graduallyafter it is being turned OFF. Again the unit should he heated up to theoperating temperature required for the scheduled turn ON. It is from thisoperating temperature that energy is expended further to hring the unit online. The time required to tum ON the unit and bring it on line if it wasinitially turned OFF or at a temperature equal to the normal value is calledcold start hours.(vi)Fuel ConstraintsA system in which some units have limited fuel or else have theconstraints that require them to bum a specified amount of f'uel in a giventime, presents the most challenging prohlem to unit commitment problem asfuel constraints.(vii) Economic ConstraintsThe temperature and pressure of a thermal unit need to he variedgradually. So a certain amount of energy is expended to bring the unit online or to shut down the unit. This energy does not affect the MW generationof the unit and is brought into unit commitment problem as stm up andshutdown costs, respectively.

Hot Start-Up Cost,It is the start-up cost involved to bring the unit on line if it is initially atoperating temperature required for scheduled turn on.Cold Start-Up Cost.It is the start-up cost involved to bring the unit on line if it is initiallyturned OFF or at a temperature close to normal value.3.3. PROPOSED GA-PSO BASED METHODThe UC problem can be considered as two linked optimization subproblems.namely the unit-scheduling problem. and the economic load dispatch problem. Thischapter proposes a method to integrate genetic algorithm with the particle swamoptimi~ation method to solve the UC problem. 'lhe GA algorithm is used to solve theunit commitment problem for obtaining the ON-OFF status of the generating unitsand the PSO method is used to solve the ELD pmhlem for obtaining the minimumproduction cost. The generating unit's combination with the lowest associatedgeneration cost will be an optimal solution.3.3.1. lmplementation of GA for LIC Solutionfirllows:The details of the implementation of GA components are summarized here as3.3.1.1. Coding of SolutionThe solution in the unit commitment problem is represented by a binary matrix(IJ) of dimension t x n. The proposed method for coding is a mix of binary anddecimal numbers. Fach column vector in the solution matrix (which is the operationschedule of one unit) of length T is converted to its equivalent decimal number. Thesolution matrix is then converted into one row vector (chromosome) of N decimalnumbers (Ul, U2 ..Un), each represents the schedule of one unit. The numbers U1,

2. ... lln are integers ranging h m 0 to 2" - I. Accordingl?. a plpulation of siteYI'OP is randoml) generated in a matrix nPOP . n.4.3.1.2. Fitness functionAs me arr aIma!s generating ti.ahihle aolutirmrdh~,t\\ecn LO chro1~iso1iic~1 in then h111a1q litrni Appl! ing mindom crosso\c.r operatorI,) parcnts X and Y. and (rhtalning olllprillg /. ~ orLs ah \I~oMI~ in I ig, 3 ICROSSOVERIFig. 3.1. Window crossover(i)(li)(iii)Set parent X to have fitness (X);- fitness (Y)Sample I from uniform (0.T). u from uniform (0.N). u from uniform(1.N-n). and h from uniform ( 1 .-1--I ) distributions.Define a uindom W,. of dimension lu. with left upper comer lu,l] andright lower comer [u+w.h+l].

(iv) Define a same window W2 in the TI Y.(v)Now, swap the entire bits of windows WI and W2 and generate twoofffsprings.3.3.1.4. MutationMutation operation is performed by randomly selecting a chromosome with aspecified probability. The selected chromosome is decoded to its binary equivalent.l'hen the unit number and time period are randomly selected and the rule of mutationis applied to reverse the status of units keeping the feasibility of the constraints [loo].3.3.2. Implementation PSO for ELD SolutionIn this approach. a quick solution to solve a constrained ELD problem usingI'SO is developed as in section 2.2 to obtain a high quality solution. Ihe PSOalgorithm is utilired mainly to determine the optimal allocation of power among thegenerating units. which arc scheduled to operate at the specific period. thusminimizing the total production cost.3.3.3. Step-by-step AlgorithmThe searching procedures of the proposed GA-PSO method are as show below.1. Initialize randomly the individuals of the population according to the limit ofeach unit including dimensions, searching points and velocities. This includesthe initial schedule of binary bits 0 and 1 analogous to the chromosomes of therandomly generated population.2. These schedules are tested for solution feasibility (generation >load) usingfitness value function where infeasible strings are eliminated and new randomschedules are generated.

3. For each chromosome economic load dispatch is performal using panicleswarm optimization method for each hour so that total load demand issatisfied. This represents the satisfaction of equality constraint (3.3)4. Genetic operation which involves selection, window crossover and mutation isthen performed on the population of chromosomes.5. The above population is checked for the satisfaction of the feasibility such asminimum up time, minimum down time and equality constraints for anyviolation in the above population. The hit positions are corrected and thesolution will be converted into feasible solution.6. If the maximum iteration number is reached. the process is stopped. Otherwisego to step 3.The procedure may be terminated once the maximum iteration count isreached or when there is no improvement in the hest solution ohtained so far in thewhole run. In this approach. the procedure is terminated when there is noimprovement in the best solution for a prespecified number of iterations.3.4. SIMULATION RESULTSThe efficiency of the proposed algorithm has been demonstrated by solvingthe unit commitment problem using a 24-hour scheduling horizon for a bench-test unitcommitment problem consisting of 10 units. For simplicity, the shutdown cost of thegenerators has been taken as zero for every unit.The PSO method is sensitive to the tuning of weights and parameters..4ccording to the experience of many experiments, the simulation parameters areshown in Table 3.1. The generators and demand data for this test system 1971 areshown in Tables 3.2 and 3.3, respectively.Table 3.4 shows the best combination of the scheduled units in the initialpopulation. Table 3.5 gives the details of generalors output power for each hour after

performing economic load dispatch for the initial population. The total productioncost during the scheduling period is $596,864.81The final commitment schedule of the GA-PSO based technique is shown inTable 3.6. The results show that the final production cost using the proposedtechnique is $ 564,620.29 a. shown in Table 3.7. The final production cost obtainedusing the GA-PSO technique is compared with the production cost of $565.804oblained using BPSO technique. Fig. 3.2 shows the convergence characteristics of a10-unit system for 10,000 iterations during GA-PSO processing. The comparison ofthe total production cost is shown in Table 3.8 for the 10-unit system. From thecomparison. it is found that the GA-PSO method has less production cost whencompared with the conventional and BPSO search methodsTable 3.1. Parameters used in GA-PSO method - 10-unit system-- -ParametervalueINumber of chromosomesChromosome s~zeNumber of generationsInertia weight factor (W)-Mutation probab~l~ty10-.24 (hours) x Number of generators ( I 0)10000Wm,=09and W,,,=04cl= 2 0 and c2= 2 0-- a0 650 001--- --.

Tnhle 3.2. C'ocl coefficients - IO-unit systemTable 3.3. Daily generation - 10-unit system

GenerationsFig. 3.2. GA-PSO convergence characteristics - 10-unit systemTable 3.8. Comparison of solution qualityMethodIProposed GA-PSO564.620.29

3.5. CONCLUSIONIn this approach a hybrid GA-PSO based algorithm solves the unitcommitment problem. The proposed algorithm integrates the main features of themost commonly used swarm intelligence methods such as GA and PSO for solvingcombinational optimization problems. The algorithm is based mainly on the PSOwhereas the GA method is used to generate new members in the population to guidethe search towards the optimal solution. The use of genetic scheme improves theperformance of coding the combination of units and to arrange the ON I OFF status ofthe units. PSO is used for power output estimation and to locate the global optimalsolution by fine tuning the search process. The implementation of the proposedmethod is demonstrated using a test system of 10 units. The results proved theeffectiveness of the algorithm in solving the UC problem with a reduced productioncost.

CHAPTER 4PSO BASED REAL POWER LOSS MINIMIZATION ANDVOLTAGE STABILITY ENHANCEMENT1.1. INTRODUCTIONOne of the important operating requirements of'a reliable power system is tomaintain the voltage within the permissible ranges to ensure a high quality of'customer service. The control of reactive power and voltage control problems hasgained importance to ensure a reliable quality of power supply with minimum lossesin the power system. Conventional search routines to solve the optimal reactive powercontrol have the common defect of being caught at local minima. Therefore. a newmethod of achieving a reasonahle voltage profile for economic and stable operation ofa power system is the need of the hour.An analysis of MW and MVAR management for the improvement ofeconomical dispatch by using panicipation factors has ken derived from the criticalclgenvectors of the jacohian matrix 11021. A new model Ibr optimal reactive pnwerllow has been designed by the predictor corrector primal dual interior point methcd11031. A bacteria foraging technique ha? been implemented for minimizing loss.taking voltage stahility into account [68]. A recent work on the stability index hasheen carried out by the use of tellegen's theorem 110.51. Other works on the stahilityindex have included preventive control of voltage stability using a new voltagestability index [106]. Many novel methods have been employed for this voltagestability control such as the effect of load tap changers in emergency and preventivevoltage stability control [107]. Nonlinear optimization techniques have been used forvoltage stability analysis by fast computation of voltage stability security margins11081. Other applications of the nonlinear programming have included congestionmanagement problem ensuring voltage stability [109]. Recent research works on thereal power loss minimization have been carried out by the use of various evolutionarytechniques. The real power loss minimization has been mainly carried out to meet outthe improvement of the voltage profile by GA technique [I 101. The application of GA

to the corrective control of voltage and reactive power has been ~nvest~gatcd 1691.Other works on voltage control have included the application of pseudo gradlentevolutionary programming to the optimal voltage control for power system stabll~ty[112], voltage and reactive power estimation for contingency analys~s usingscns~tivities [I 131 and hence global nonlinear control has been implemented.This PSO based algorithm uses optimum settings of Automatic VoltageRegulator (AVR), On Load Tap Changer (OLTC) and hence finds the mlnltnumnumber of Reactlve Power Compensation Equipments (RPCE) to hc connected In thesystem. This swarm intelligence technique differs in the sense that 11 ~dcnl~lics theglobal appropriate values in its bid to offer the best possible voltage profile. The IdeaIS to search the global optimum settlngs of AVR values, OLTC tap posltlons and thenumher of RPCE's to be connected in order to minimize the real power losses therehy~mproving the voltage stability of the power system. The voltage stabil~ty assessment1s performed using a line voltage stability index. The particle swarm opt~mizat~ontechn~que using MATLAB for d~fferent systems such as Standard-5, IEEE-14. IEEE-30. IEEE-57, IEEE-I18 bus systems, an lnd~an util~ly system such as Ncyvel~Thermal Power-Station (NTPS) system and Puducherry bus system have been caniedout and the performance evaluation is presented. Thc results ohtained uslng PSO 1sfound to provide mlnimum real power loss when compared to Newton-Raphson basedload flow method and genetic algonthm approach, thus improving the voltagestability of the power system.4.2. PROBLEM DESCRIPTIONThe losses that occur in a power system have to be minimized in order toenhance its overall performance. In the proposed work, a new algorithm that attcmptsto minimize the real power losses, with a view to improve the voltage stability of thesystem has been developed.4.2.1. Objective FunctionThe objective function is given 1641 as follows

Minimize f,(x, y) = loss, (4 I)5-1The above function 1s an equality function, where is the control funct~ongoverning both the control parameters x and y to minimize the real power loss.wherenl is the number of branches.x is the cont~nuous variables. (AVR values),y is the dlscrete vanables. (OLTC and SC values).Loss, 1s the power loss (PI) at branch i.Subject to the following constraints(i) Voltage constraint in whlch voltagc magn~tude at each node should he w~th~ntheir permissible range [64] (V,,,, and V,,,)v,,, (b) 5 V(b) 5 VnUz (b)(4.2)where V,,.(b), V,,(b) are the minimum and maximum limits for the voltagcmagnitude, V(b) at each nodc, b.(ii) Reactive power constraint In whlch reactive power at each nodc must hewithin their permissible ranges [MI (Q,,, and Q,)Q,. (b) Q(h) 5- Q,,(b) (4.3)where Q,,,(b), Q,,(b) are the rn~nimum and maximum llmits for rcactlve(iii) Voltage stability,power, Q(b) at each node, b.A strategy should be provided to keep track of the condition of thelines in the target power system within an acceptable limit. It is achieved by thecomputation of the line stability index [115].LS. 5 1 (4.4)where LS, is the line stability index at i~ line or branch.The wntmt variables in this problem are:a) AVR operating values (continuous variable)

) OLTC tap position (discrete variable)c) The number of RPCE (discrete variable)The above variables are treated in load flow calculation as follows: AVRoperating values are formulated from the voltage specification values, OLTC tapposition are generated with the help of tap ratio to each tap position and the number ofreactive power compensation equipment to be connected is calculated homcorresponding susceptance values.4.3. VOLTAGE STABILITY ASSESSMENTThe voltage collapse prediction [114] methodology has been presented hasedon line voltage stability index [115]. It is predicted hy estimating the load flowsolution and then calculating the line voltage stability index. Ilence. the I~nes whichare in stressed conditions can be easily identified. This information can he used as abasic tool for security monitoring [I 161 ofthe system.4.3.1. Line Voltage stability indexMost of the indices developed are system-based-or based on bus orientation.'There has not been much research in case of static voltage stability [I 171 1 l lR]assessment via line based voltage stability index.In this method, an effective procedure for voltage stability assessment(nearness of the operating point to voltage collapse point) using the exact line voltagestahility index is developed [l IS]. The developed index incorporates correctly theeffect of real and reactive power increase scenario in any direction. The mathematicalformulation has been given in Appendix-E. The line stability index values arecomputed by incorporating the following equation.

2B JmLS, =V: B -- - B2~p,cos(f4-%)-2 --Qin(fj,-q)A5 Iwhere LS, is termed as voltage stability index of the i" line,P,. Q, are the real and reactive power at receiving end. respectively. in p.uVK is the voltage magnitude at sending end in p.u.A La1 and B LPI are transmission line constants.As long as the above index is less than unity, the system is stable. L.S, istermed as voltage stability index oithe line. At collapse point. the value of LS, will heunity. Based on voltage stability indices, voltage collapse can be accurately predicted.'l'he lines having high value of the index can he predicted as the critical lines. whichcontribute to voltage collapse.4.4. ALGORITHM OF THE PROPOSED METHOD1. Initial search points and velocities are randomly generated for each of thethree variables between their upper and lower hounds.2. Pl. for each set (one value of AVK, OLI'C and SC') of particles is evaluatedbased on the fitness function. If the constraints are violated. penalty is added.3. Assign the particle's position to pbest position and fitness to pbest fitness.Identify the best among the pbest's as the gbest.4. New velocities and new search points (directions) are formulated using theabove equations (2.7) and (2.8). respectively.If V2Vm" then V = VM"If V

4.5. SIMULATION RESULTSThe proposed algorithm is applied to Sample 5 bus systems. lEEE -14 bus. 30bus, 57 bus 118 bus systems [147], Indian Utility-23 bus systems and lndian Utility-17 bus systems whose data have been given in Appendix-F. 4. -A, -H, -I. -J. Thesimulation parameters considered for the above test cases are shown in Table 4.1.Table 4.1. Parameters used in PSO method - Standard-5, IEEE-14, -30. -57.-118, Indian utility -23 and -17 bus systemsParameterPopulat~on sizeNumber of iterationsChosen ValueInertia weight factor W,, = 0.9 and W,,. = 0.4Velocity limitsvma'= I and vmIn = - 1Acceleration coefficients 1 c, = c2 = 2 . 0 ~ - 14.5.1. Standard -5 bus systemA standard-5 bus system has 7 branches. 2 generator buses and the rest are theload buses.a. Continuous AVK operating values of generators and synchronouscompensators (SCs) are connected at node 2. The upper and lower boundsof AVR are set to 0.9 and 1 .I [p.u].b. Discrete tap positions of transformer are assumed between nodes 4 and 5.The transformer is assumed to have 20 tap positions.c. A discrete synchronous compensator (SC) is installed at node 5. The nodeis assumed to have value between 0.954 and 1.038 [p.u] SC.4.5.2. Standard lEEE systemsIn case of standard systems, to forecast the advantages of PSO on them, theapproach presents an analysis of a set of IEEE systems - IEEE-14, -30, -57 and -1 18

us systems as numerical examples. which enumerate the favorable effects of PSO onthem. The programming conditions for IEEE-14, -30 end -57 bus systems arc asfollows:IEEE-14 bus systemContinuous AVR operating value is given at node 2. Discrete tap position oftransformers is provided at end nodes 6. 7 and 9. Node 9 is assumed to have threenumbers of 0.19 [p.u] SC.IEEE-30 bus systemDiscrete lap positions of transformers are provided at end nodes 9. 10. 12 and27. Discrete Synchronous Compensator (SCs) are inscalled at nodes 10 and 21. Node10 is assumed to have three numbers of 0.19 [pa) SC and node 24 has three numbersof 0.043 [p.u] SC.IEEE-57 bus systemDiscrete tap positions of transformers are provided at end nodes 18. 18.20.25.25. 26, 29, 32. 41, 43. 45, 46, 49, 51, 55, 56 and 57. Node 18 is assumed to have threenumbers of 0.10 [p.u] SC, node 25 has three numbers of 0.059 Ip.u] SC' and node 53has three 0.063 [p.u] SC. The analysis is camed out on P-IV. 700 MI-lz system in aMATLAB environment.4.5.3. lndian utility systemsThe PSO method has been applied to two real time systems such as the lndianlltility (IUFNeyveli Thermal Power Slation (N'I 1's)- 23 bus system and lndian utility(IU) Pondicherry-17 bus system.lndiau Utility nu)-Neyveli Thermal Power Station (NTPS)-23 bussystemThe programming conditions for this real time system are as follows. TheNLC bus system consists of 23 buses and 22 lines. There are three generator buses,one slack bus and the remaining are load buses.

The continuous AVR operating values are given at nodes I. 2, 3 and 4.Discrete tap positions of transformers are provided at end nodes 5, 6, 7. 8. 9, 10. 11.12, 13. 14, 15, 16. 17, 18, 19,20,21,22, 23.Indian Utility (IU) Puducherry-I7 bus systemPuducheny bus system consists of 17 buses and 21 lines. It includes agenerator bus and the rest of them are load buses.The real power losses. PI ohtained for the test systems using conventionalN-R. GA and proposed PSO based approach are given in Table 4.2. lhe optimalsettings of the control variables for varioug systems are also given in 'l'ahles 4.3. 4.4and 4.5.Table 4.2. Comparison of real power losses with different methodsStandard-5 bus.-IEEE-14 bus 12.0392 1 1.7971-- 9.6277-IEEE-30 bus 17.3790 17.2215 16.8264IEEE-57 bus 27.60 13 27.5595 27.5345IU-17 bus j 2.3674 44TTTTTm-~~1.99IU-23 bus 14.8417 34.5928 6.6593Table 4.3. Optimal control variables - IEEE-14 bus system/ Bus/ No.IEEE-14 Bus SystemAVR 1p.u.j OLTC 1p.u.j I = Ip.u.1

* 1 : 0.19 Lp.u] 1 number of SC is connected for an IEEE-14 bus systemTabk 4.4. Optimal control variabla - IEEE-30 and -57 bus systems560.9971 -570.9958 -*3: 0.19 Lp.u] x 3 numhers of SCs are connected and 0.043 [p.u] x 3 numbers ofSCs are connected for IEEE-30 bus system.'2: 0.10 [p.u] ': 2 numbers of SCs are connected, 0.059 [p.u] x 2 numbers of SCsand0.063 Lp.u] x 2 numbers of SCs are connected for IEEE-57 bus system.

Table 4.5. Optimal control variables - IU- NTPSU bus systemBus No.IU-23 Bus SystemAVR 1p.u.jOLTC 1p.u.l10.9798 -The above analysis proves that PSO-based strategy is better than the earlierconventional techniques.

-System with Bus LocationFig. 4.1. Variation of 8oltnge profile for diffirrnt optimkntion methodsAn anallhi5 of the \~llagc prolilc ohtilined using thc thrcc niclhoda li~r 11.1 1.-31. -57 and -1 I X huh .;>\tern\ 15 \ho\rn in I I; 4.1 Sincr \oll:~gr prolilc :II the huwsf 11"' hw in I1 I I -30. 53" h~15 ill 11.1 I -i7 i111d (31"h115 III I1 I I -I IN) \IIo\\ a great\.lrlatlon :many the thrcc methods. the) \\ere \pcc~licallq selected li~r the plc~t. lh~i~aphtghlights the s~gnificancc 01. I'SO ha\cd ilpproach in the sense. thc htratcg)ollcrh a hiyhcr ~oltayr. pn~lile compared to the other I\\() rncth~dbTnhle 4.6. Solution of stnhilib assessment using line voltnge stability index- IEKE-I4 hus nnd It!-NTPS-23 hus systems

As discussed earlier, if the index value IS less than unity, the system is said tohe stable. From the above Table 4.6. there are 20 lines in case of 1EEI'-14 hus and 22lines in case of IU-NTPS-23 bus system and each of their index values does notexceed 1 which indicates that all the lines are secure and thus both the system remainin stable conditionFig. 4.2 shows the typical convergence characteristic of IEEE-I 18 bus system.Here the problem converges at about 7oLh iteration. in which the best PI value isobtained.

Fig. 4.2. PSO-based conveqence chnrncteristics - IEEE-118 bus system1.6. CONCLUSIONThe application of the PSO technique for the problem of minimization of realpower loss taking voltage stability into account has been detailed. The proposcdmethod also determines a control strategy with continuous and discrete variables suchas AVR operating values. OL'I'C tap positions and the number of RI'CE. The featuresofthis technique are summarized as follows:i) The power loss is reduced drastically.ii) The feasibility of the proposed PSO method for the prohlem is demonstratedfor standard IEEE systems and IU real time systems with promising results.iii) The algorithm has paved the way for realizing an acceptable voltage profile,by reducing the real power losses.iv) The technique is fairly simple, in thc sense, that it does not involve anycomplicated procedures.

CHAPTER 5SWARM INTELLIGENCE BASED OPTIMIZATlON OFDISTRIBUTED GENERATION CAPACITY FOR POWERQUALITY IMPROVEMENT5.1. INTRODUCTIONDistributed power generation is a small-scale power generation technologythat provides electric power at a site closer to the customers than the centralgenerating stations. Distributed generation provides a multitude of services to utilitiesand consumers, including standby generation, peak chopping capability. base loadgmeration. Investments in distributed generation enhance onsite efficiency andprovide environmental benefits, particularly in combined heat and power applications.A multitude of events have created a new environment for the electric powerinkstructure. The key element of this new environment is to build and operatese\eral DG units near load centers instead of expanding the central-station powerplants located far away from customers to meet increasing load demand. Distributedgeneration technologies can enhance the eff~ciency, reliability, and operationalhenetits of the distribution system. A distributed power unit can he connected directlyto the consumer or to a utility's transmission or distribution system to provide peakingsen ices.DG can be powered by both conventional and renewable energy sources (711.Several DG options are fast becoming economically viable [I 19-1271. Technologiesthat utilize conventional energy sources includes gas turbines. micro twbines and ICengines. Currently, the ones that show promises for DG applications are wind electricconversion systems (WECS), geothermal systems, solar-thermalkelectric systems,photovoltaic systems (PV) and fuel cells. T. Hoff et al. [I281 have discussed thebenefits of DG by evaluating and quantifying in terms of capacity credit, energy valueand energy cost saving. The effects of improvement in voltage profile and lossreduction were not wnsidered in the method. Y.Z. Hegazy et al. [I 291 have presenteda Monte Carldased method for the adequacy assessment of distributed generation

systems. K0enJ.P. Macken et al. 11301 have presented solutions to prevent sensitiveequations from disruptive operation by making use of DG in the presence of voltagedips. Aleksander Pregelj et al. [I311 have demonstrated a combination of clusteringtechniques and convex hull algorithm for analysis of large sets in renewabledistributed generation. Caisheng Wang et al. [I321 have presented noniterativeanalytical approaches to determine the optimal location for placing DG in both radialand networked systems to minimize power losses. All the above methods [74.129-1321 are mathematically modeled and hence are found to be complex in its approach.DG. Victor H. Mendez Quezada et al. [I331 have presented an approach tocompute annual energy losses when different penetration and concentration levels ofDG are connected to a distributed network. The method also identifies that when DGunits are more dispersed along the network feeders, the expected higher losses can bereduced upto a particular DG capacity beyond which the loss increases. This idea isused in the proposed method for optimizing the DG capacity corresponding tom~nimwn power loss and P. Chiradeja [I 341 has quantified the benefit of reduced lineloss in a radial distribution feeder with concentrated load. R. Ramakumar et a]. 1711have proposed an approach to enumerate the various power quality indices in terms of\joltage profile. line-loss reduction and environment impact reduction.The proposed work finds out the optimal value of the DG capacity to beconnected to the existing system using Particle Swarm Optimization (PSO) techniquethereby maximizing the power quality. Benefits of employing DG are analysed usingVoltage Profile Improvement lndex (VPII) and Line Loss Reduction lndex (LLRI).rhe Line Voltage Stability lndex (LVSI) obtained by performing a conventionalNwton-Raphson load flow method calculates accurately the proximity of theoperating point to the voltage collapse point. The calculated Line Voltage Stabilityindex using the proposed PSO based method exactly finds out the critical lines as thatof the conventional method and hence validates the significance of the proposedmethod. The optimum value of the DG obtained increases the maximum loadability ofthe system. The proposed method is tested on a standard IEEE-30 bus system. Themethod has a potential to be a tool for identifying the best location and rating of a DGto be installed for maximizing power quality in an electrical power system.

5.2. APPROACH TO QUANTIFY THE BENEFITS OF DCA set of indices is proposed to quantlfy some of the technical benefits of 1Xi.They are VPII. LLRl and LVSl5.2.1. Voltage Profile lmprovement IndexThe inclusion of distributed generatron results In Improved voltage profile at\anous buses of the power system and to malntaln the voltage at the customer sltewlthtn the operating range. As the current flows through the distnbutlon Ilne. voltagcdrop occurs at the customer tcrmlnals. Adopt~ng DG Into the exlsting power systcnlcan provide a portlon of the real and reactlve power to the load This helps 811decreasing the current along the portlon of the distnbut~on hne. Hence the systcmvoltagc profile can be improved. The voltage profile lmprovement tndex quantifiesthe lmprovement In the Voltage Profile (VP) with the inclus~on of DC; 171). The\.oltage profile improvement 1ndex is defined as the ratlo of voltage profile of the loadhuses In the system w~th DG to that of voltage profile of the load buses ofthe systemw~thout DG [71].where, VP wDc, VP W~IX, are the measures of the voltage profile of the systemwith DG and without DG, respectively. The general expression for VP (711 1s givenas.

%here. V,, is the voltage magnitude at bus I In p.u.. L, IS the load rcpmented as,,,mplex bus power at bus i in p.u.. K, 1s the weighing factor for bus i. and N 1s thctotal number of load buses in the dlstnbutton system.The weighting factors are chosen based on the importance and cnt~eality ofdtfferent loads. As defined, the expression for VP provldes an opportunity to quanttfy:md aggregate the importance. amounts, and the voltage levels at which loads arebang supplied at the vanous load buses In the system. It allows the posstbil~ty of slow-load bus with important load to have a strong Impact This expression should beused only aAer making sure that the vc~ltages at all the load buses are withln allowablemlntrnum and maximum lim~ts, typically between 0.95 p.u, and 1.05 p.u. Theuclght~ng factors are chosen based on the nnportance and crtt~cal~ty of the d~flcrentIoclds Starting with a set of equal wcightlng factors, rnodlfications can he made and.h;lsrd on an analysis of the results, the set that will lead to the most acceptable voltagprofile on a systm-wide basis can be selected It should he noted that IF all the loadhuscs arc equally weighted. the value of K, 1s 1721 given asIn this case all the load buses are given equal irnporlancc. In realtty, DG canhc tnstalled almost anywhere in the system. In general, the htghest value of VPll~rnpl~es the best location for installing DG in terms of improving voltage profile.Therefore, VPIl can be used to select the best location for DG.Rased on this definition, the following attributes are,i) VPll c 1; DG has not been beneficial.ii) VPIl = 1; DG has no Impact on the system voltage profile.iii) VPU >I; DG has improved the voltage profile of the system.

5.2.2. Line loss redoction indexAnother major benefit offered hy tnstallatton of DG is the reduction inclectncal line losses [71]. Electncal ltne losses occur when current flows throughtransmission and distribution systems. The loss can he significant under heavy loadc~ndtttons. The magnitude of loss depends on the amount of current flow and linercs~stance. Therefore. electrical line losses can be decreased by reducing etther thel111current or line resistance or both. By installtng DG, ltne currents can he reduced.thu5 helping to reduce electrical ltne losses The line loss reduction Index (LLRI) isilcfined as the ratio of total line losses in the system wtth I)G to the total line losses InIIK system without DG and it IS expressed 1711 as,where, LLwmj is the total Itne losses in the system with the employment ofI)(; and LLwdrx; is the total line losses In the syst~m wlthout DG and 11 can bedeterrnlned hy summing all the bus powers ohtained aner periormlng Newton-Kaphson load flow solut~ons.f3ascd on this definition, the following attributes are,i) LLRl c 1, DG has reduced electncal llne losses,ii) LLRl = I; DG has no Impact on system llne losses.iii) LLRl > 1; DG has caused more electrical line losses.This index can be used to identify the best location to tnstall ffi to maximlzcthe line loss reduction. The minimum value of LLRl corresponds to the best DGlocatton scenano in terms of line loss reduction.5.2.3. Line Voltage Stmbility Index (LVSI)To assess whether a normal operating state is secure or not the owline loadflow calculation and the stability constraints are evaluated. Basal on these stability

,muring wnshaints the security assessment is canicd out. The method utilizes linestability index to monitor the system stability. Based on IIK stability indices of lines.voltage collapse can be accurately predicted. As long as the stability index L, remainsless than 1. the system is stable. When this index exceeds the value I. the wholesystem loses its stability and voltage collapse occurs. Using this technique ofcalculating line stability index. the status of all the connected lines of the network ismonitored and identification of the lines which are in stressed condition is performed1115).The line stability index is elaborately discussed in Appendix E and is given by.where.I.S, is termed as voltage stability index of the i" line.P.,, Q, are the receiving end real and reactive power, respccti\,ely. inP.U.,VI. is the sending end voltage magnitude in p.u...4 Lu, and H LP, are transmission line constants.5.3. PROBLEM DESCRIPTIONIn order to evaluate and quantify the henefits of distributed gencration,\uitahle mathematical models must be employed along with distribution systemmodels and power flow calculations are carried out to arrive at indices of henefits.among the many benefits three major ones arc considered: voltage profileImprovement index, line loss reduction index and line voltage stability index.

53.1. Objective FunctionThc proposed work aims at minimizing the combined objective functiondesigned to reduce power loss and also improve system voltage profile for various\dues of distributed generations. The main objective function is definedHminf = P, +xkp(l-~n)'. (5.7)where. & is the penalty factor of bus voltages and is heuristically taken as 1.PI,,, is the real power loss ohtained from the load flow solution at the basecase andVp is the voltage profile of the huses.The values of the I>G are taken as the particles to he optimiz.cd for obtaining arn~nimum value of ohjective function. The upper and lower values of I>(; ure fixedhased on the availability of the power generation at the site.5.4. ALGORITHM OF THE PROPOSED METHODI hc sequential steps arc as follows:I. Randomly generate the panicles value between upper and lower limits ol I)(;capacity.2 Assign the initial particle value as the pbest values.X Compute the objective function of each particle with its phest and the hestamong the pbest is gbest.4. Change the velocity and position ofthe panicle using (2.7) and (2.8).If V>Vmu then V = VmIf V

6. Compare the best current fitness evaluation with the population's gbest. If thecuma value is bmer than the gbesL then rem gbest to cumnt besi positionand fitness value.7. Repeat Step 4 to Step 6 until the convergence criterion of maximum numberof evaluations are met.8. Corresponding to optimal DG. calculate line suability index values.5.5. SIMULATION RESULTSThe study has been conducted on an IEEE-30 bus system whose data havehrrn given in Appendix-A. The buses having less voltage profile are identified aReryrforrning NeWon-Raphsnn load flow analysis and are chosen as the locations for~hc L)Gs to be installed. tlere the bus numbers 30. 26. 7 and 29 have less voltageprofiles at the base case and hence were chosen as locations for DG installations.lehlc 5.1.Case 1 : I>G located at Bus 30.Case 2: DO located at Bus 26.Case 3: Dti located at Bus 7.Case 4: 50 % of DG located at Bus 30 and Bus 26.Case 5: 50% of DC; located at Bus 7 and I3us 29.The simulation parameters consid~~ed for the ahove test cases arc shown inTable 5.1. Parameters used in PSO metbod - IEEE-30 bus system.--ParameterValue-Population sire 20- --Number of iterations100-- - P . . ~Inertia weight factorW,, = 0.9 and W, = 0.4--Velocity limitsVm" = 0.5 and Vmm= 0.05--Acceleration coefficientsc, = c2 =2.0-5.5.1. Results of lmprovemeot in Voltage Profile and Line h s ReductionTables 5.2 and 5.3 indicate ha^ for the various cases considered the values ofthe voltage profile of the system have improved considerably by connecting a DG of

various capacities. The voltage profile of the base case was calculated to be?.9441 p.u. When a DG rating of 0.2 p.u. and 0.3 p.u. were connected for cases 1 to 5.the voltage profile of the system has improved which clearly indicates the need of aDG. It should be noted that the voltages at every bus before employing a W (basecase) was maintained within 5% of the reference voltage (1 pa.). Therefore animprovement of about 1-1.8 % indicates a significant impact on the voltage prolile ofthe load buses. Fig. 5.1 shows variation of improvement in voltage profile at bus 30for different DG ratings.Table 5.2. Voltage profile improvement results for a DC rating 0.2 p.u.r- . - - -Rase case 2 9441*----Case 1 2.95 10 1.0135 1.35Case 2 2.9492 1.0129~1.29I Case 3 1 2.9458 1 1.0117 1 1.17 1Case 4 7 9509 1 01151Case 5 2.9479 1.0124-FITable 5.3. Voltage profile improvement results for a DG rating 0.3 p.u.CasesBase caseCase 1Case 2Case 3Case 4I Case 5VP (P.u.)291172 95312 95032 94662 95342 9495M; rating 0.3 p.u.VPll Improvemen1 %-- --- .101421 42I 01331 33101201.20101431 43101301 30

Dc (MW)Fig. 5.1. Voltage profile improvement results with different I>ti ratinps]-he reduction in line losses is evident after connecting IXi as shown in l'ahlesi 4 and 5.5. It indicates the reduction in line losses with the installation of I)(; for(anous cases. lhe line loss Ibr the hase case without I)(; installation is calculated byload flow solutions and is found to he 0.2836 p.u. for DCis of0.2 p.u. and 0.3 p.u. The\alucs of line loss considerably reduce as indicated in Tables 5.4 and 5.5. 'Ihewrccntage of line-loss reduction is indicated hy means of IdL.RI and a maximumreduction of 30.93% is obtained in Case 1 for a I>G of 0.2 p.u and a reduction of41.94 % for a DG of 0.3 p.u., respectively.I-Table 5.4. Line loss reduction results for a LK rating 0.2 p.u.DC rating of 03 p.u.CasesLine Loss (p.u) LLRl Reduction %Base case0.2836 - -Case l0 1958 0 690730 93-

~ .-0.2072 0.7309 26.910.2039 0.7193C---------Case 4 0.2007 0.6780 32.2Case 5 0.1983 0.6935 30.65i-Table 5.5. Line-loss reduction results forD(; rating 0.3 P.U.Case l 0.1750 0.6170. -Case 2 0.1932 0.6812 31.88e---- -care 3 T ~ r 7 i i - 7 6 - i ~ I ~ 7966 1-Case 45.5.2. Comparison of Line Voltage Stability Index of Conventional andProposed MethodsWhen the DG capacity was randomly generated hetween 0.05 p.u and 0.5 p.u..the optimum value of a DCi to be connected for maximizing power quality was hundto he 0.45 p.u. using the proposed PSO method. The line volwe stahility index of themost critical line. 7" line, was found to be 0.7764 using conventional Newon-Kaphson load flow method at a DG value of 0.45 p.u. and was found to be exactly them e fur the proposed PSO-based line voltage stability index. The line voltagestahility index values for various lines using the conventional load flow method andthe proposed PSO-based method for the optimum DG capacity of 0.45 p.u. are foundto be almost identical and are shown in Table 5.6.

Table 5.6. Comparison ofthe conventional Newton-Raphaon and proposcdmethods in cnrpnting line volbge sbhility indexConventional MethodProp04 PSO MetbodLine Numher Line Index I Line Number / Line IndexThe effectiveness of the proposed method is evident liom the correctness inthe determination of the critical lines at various loading conditions ahove the hasccase for an IEEE-30 hus system as that of the conventional method as indicated inl ahle 5.7. For different loading conditions up to a critical loading of 273.942%. the7Ih and 13* lines were consistently critical as calculated by conventional Ne\r?on--Kaphson load flow method as well as the proposed PSO method. This validates theaccuracy of the proposed methodTable 5.7. Detection of critical lines using different methods% of Base Case Conventional Newton-Proposed PSO Methodn i h s o n Method 1 -‘:~---1 273.9420.8057 1 7m-iine -a7rLpe----\ ---ip

ITable 5.8. Solution of improvement in system loadabilityMaximum After Including% of lncmasc inDG at Bus Loadability a DG of 03Loadability(in p.u.)p. IL5.6. CONCLUSIONIhe proposed work has presented a swarm intelligence based approach toquantify some of the benefits of installing a IXi. nanlely. voltage profile~mprovement. line loss reduction and improvement of system loadah~lity. lheproposed method is applied to an IEFE-30 hus system and it clearlj indicutes that theI)(;can improve the voltage profile and rcducc electrical line losses cmd improve thel~ne \ohage stability. Both ratings and locations of DCi have to he considered together\er! carefully to capture the maximum henefits of Mi. lhc capability of I'SO is tomaximize the power quality hj optimizing the DG capacity.

CHAPTER 6PARTICLE SWARM OPTIMIZATION BASED DYNAMICVOLTAGE RESTORER6.1. INTRODUCTIONThe recent grouzh in the use of power electronics has cauxd a greaterdHareness on power quality. Voltage sags. swells and harmonics can cause ancqulpment to fail or shut down. and also create huge current imhalanccs ujhich couldblow fuses or trip the breakers. These etyects can he vey expensive fbr the customer.ranging from minor quality variations to production downtime and equipmentdamage. Utilities are interested in keeping their customers satisfied and also keepingthem on-line and drawing kilowatts. creating more revenue for the utility. All of thisInterest ha? resulted in a variety of' devices designed for mitigating power disturbances\uch as voltage sags. One class ol.the device is the Dynamic Voltage Restorer (INK).I. Jauch et al. 11351 have demonstrated the in-phase voltage ~njectiontechnique where the load voltage is assumed to be in-phase with the presag voltageti~r the DVR control. N.A Samara et al. 11361 have incorporated the INK into adlstrihution network and analyzed the perfkrmance of DVK for highly sensitivetndustrial loads hased on reactive power compensation. Alexander Kara el al. 11 371have presented the technical aspects of designing a dynamic voltage restorer to meettile stringent requirements of voltage dips mitigation with respect to the magnitude of\oltage dip, fault duration. permissible line voltage deviations and response time.S W. Middle Kauff et al. [I381 have proposed that almost dl voltage disturbances are%\socialed with some degree of phase shifi for .wries custom power devices. Poh('hlang Loh et al. 11391 have presented the implementation and control of a highrollage dynamic voltage restorer for compensation of utility voltages using amultilevel inverter topology. John Godsk Nielsen el al. (1401 have proposed differentDVR control methods to reduce voltage disturbances caused by voltage sap withphase jump technique for very sensitive loads. Hyosung Kim et al. 11411 haveexploited various operation modes and boundaries such as inductive operation.

capacitive operation and minimum power operation as an effective and economicsolution to overcome voltage sags. Chris fitm a al. [I421 have proposed a novelstate-space matrix method for computation of the phase shift and voltage reduction ofthe supply voltage much quicker than the fourier transform or a phase locked loop(PLL). F. Jurado et al. [I431 have proposed a neural network control strategy forprotection of sensitive loads from the effects of voltage sags.In this chapter, PSO based approach identifies the required value of phaseadvancement angle corresponding to minimum energy injection from the energystorage element of the DVR such as a capacitor or a banery. The proposed PSO basedenergy optimization method is tested using a case study for a balanced 3-phasesystem. The energy stored in the DVR after implementing PSO technique is lesserthan that of conventional in-phase voltage injection and phase advance compensationmethods.6.2. OVERVIEW OF A DYNAMIC VOLTAGE RESTORERl'he dynamic voltage restorer is a custom power device for series connectionInto a distribution line. When connected in xries between a source and a load. theDVK can control the voltage applied to the load by injecting a voltage of arbitraryamplitude. phase and harmonic content into the line. This enables the voltage seen bythe load to be compensated to a desired magnitude in the face of upstreamd~sturbances.The DVR is capable of supplying and absorbing both real and reactive power.In many cases, small disturbances can be restored through the exchange of reactivepower only. For larger disturbances, it is necessary for the DVR to supply real powerto the load. The reactive power exchanged is generated by the inverter without anyenergy storage devices. The real power exchange requires energy storage. Therefore,the DVR should be provided with a storage device in the form of a battery or acapacitor bank. When the line returns to normal following a disturbance, the storedenergy is replenished from the distribution system by the DVR.

Fig. 6.1. Scbemntie block dingrnm of power distribution systemcompensated by a DVRA schematic diagram of the DVR incorporated into a distribution network isshown in the Fig. 6.1. V, is the supply voltage, VI is the incoming supply voltagehefore compensation. VZ is the load voltage after compensation, Vdvr is the series~njected voltage of the DVR and 1 is the line current. Dynamic voltage restorerconsists of an injection transformer in which the secondary winding of the transformerIS connected in series with the distribution line. Also a voltage-source pulse widthmodulation inverter bridge is connected to the primary of the injection transformerand the energy storage device is connected at the dc-link of the inverter bridge. Theinverter bridge output is filtered in order to mitigate the switching frequencyharmonics generated in the inverter. The series injected voltage of the DVR, Vdvr. issynthesized by modulating pulse width of the inverter-hridge switches. The injectionof-an appropriate Vdvr. in the face of upstream voltage disturbances demands a certainamount of real and reactive pwer requirement from the DVR.6.3. EXISTING DVR STRATEGIESAs was recognized by many researchers. the compensation correctioncapability of the restorer concentrates to improve the voltage quality by adjusting theboltage magnitude. wave shape, and phase shift during the occurrence of voltage sag.

~xtensively used in present DVR control is the w called in-phase voltage inject~ontechnique and phase advance compensation techn~quc.6.3.1. In-Phase Voltage Injection techniqueIn this technique the load voltage is assumed to be in-phase with the presagboltage by injecting AC voltage in series with the incoming three phase network [I 351and [144]. This strategy is applied to both halanced and unbalanced voltage sags.Ilowever, this method does not take into account the phase shifl of the voltaged~sturbances. Therefore the power needed to inject from the DVR energy storage unltInto the distribution system was maximum. Hence this technique does not take intoaccount the minimization of the energy required to achieve a required voltagerestoration. For sags of long duration, this could result In poor load ride-throughcapability.The steady state active power injection from the DVR when using the in-phaseboltage injection technique is given [77] as follows.where V2 IS the balanced output voltage,I is the balanced load current,0 is the load power factor angle,V, is the source side voltage,6 is the supply voltage phase angle,the subscript j represents j" phase and j =I, 2.3.Similarly, the steady state reactive power injection from the DVR when usingthe in-phase voltage injection is given [77] as follows,

63.2. Phase Advance Compensation (PAC) techniqueThe function of the DVR shown in Fig. 6.1 is to ensure that any load voltagedisturbances can be compensated effectively and the disturbance is transparent to theload. The corresponding phasor diagram describing the electrical conditions duringthe voltage sag compensated by PAC scheme is depicted in Fig. 6.2 wherr only theaffected phase is shown for clarity. Let I. @. 6 and a represent the load current, loadpower factor angle, supply voltage phase angle, and load voltage advance angle.respectively. Unlike the in-phase voltage injection technique considered in 11351 and11441, the phase advance compensation (PAC) technique 1801 is realized hy theadjustment in load voltage advance angle a. One major advantage of the proposedscheme is that less real power needs to be injected from DVR energy storqe unit intothe distribution system. Compared to the conventional in-phase injection method, thephase advance compensation scheme permits the DVR to help the load ride throughmore severe voltage .sags. However. the advancement of load voltage advance angle uat the beginning of compensation as well as the restoration of phase angle at the endof sag must he carried out gradually in order not to disturb or interrupt the operationof sensitive loads.Fig. 6.2. Phrser diagram of power distribution system during sag

6.3.2.1. DVR Power FlowThe power flow calculation of the DVR under the phase advancecompensation technique [SO] is considered as follows.I. DVR Power FlowIf P,, and P ,respectively [SO]. thenare the input power from the source and the load power.Assume a balanced Load(IJ = 1) and a balanced output voltage (VZ, = Vl)Po, = 3 v, I Cos (Q)Let Pd, be the real power supplied by the DVR, then from (6.3) and (6.5)Pdvr = Pout - P,npdvr = 3v21c= (e) C v,, I, cm (0- m + sl) (6.6)Similarly ifQ, and Q,respectively [80], thenVIbe the input reactive power from the source and loadQ,, = x~,,l,~in(@-a+G,) (6.7)VJReactive power supplied by DVR will beQ,=Q,-Q.

From (6.6) and (6.10) it is obvious that the control of real power and reactivep)wer exchange between DVR and distribution system IS possible only with theadlustmen1 ofthe phase angle n for a given value of 6. a, V,. V2.11. Minimum Power OperationThe real power and reactive powers suppl~ed by the DVR depends on thenature of voltage disturbance expenenced as well as the dlrmt~on of the DVR injected\.oltage with reference to the presag voltage. Pdvr depends on the advance angle a for agven 6 and VI, as shown in Fig. 6.2. Based on the values of n used, the minimum~alue of P,jw can be negative. This impl~es that the real power 1s bang absorbed byL)VR However, there is no technical and economical advantage hy operating this wayduring the sag period, the DVR should be exporting energy to support the load insteadcf drawing more power from the source A negatlve Pd,, may cvcn aggravate the sagsltuatlon. A larger energy storage facility will be requ~rcd to cater for the absorbedpower for no obvlous technical advantageThe possibil~ty of operating at Pdw = 0 during sag 1s an Interesting proposition.The following analysis is therefore carried out to explore thls possibility bydetermining the corresponding value of load voltage advance angle a for such anoperation [80].Case A. Operation at Pdn= 0, h m equation (6.6)3V,ICos(O)- V,, 1, Cos (O -a + 8,) = 0 (6.1 1)v,Let X=CV,,COS(S,), Y=CV,,SI~(S,), then, following some simplemanipulations, the phase advance angle cr that corresponds to Pdvr = 0 is given by~v,Icos@)-Cv,V,, I, Cos(Q, - a + 6,)=0

(X'+Y:l0' =[i.. 1; ! ., I;]"IV,,COS(&,) + x~,,~in(fi,)Hence.where p= tan-'(y/x), it can be seen from the expressions already shown thatfor balanced sags. p = 8 and G,, is the optimum value of phaw advancement angle forminimum power operation.The necessary condition for the existence of a , is given by(X2 + Y2)O' 2 3v2cos(@) (6.22)Thus, voltage correction with zero power injection is possible only if thecondition imposed by the above equation is satisfied.

If the voltage :eg is so severe that equation (6.22) cannot he sat~sfied, then theoptimum value of load voltage advance angle a can he calculated bysettlng dPd, /d a = 0 . At this operating point, the DVR suppl~cs m~n~mum real power tothe external system to keep VI = 1 P.U.Case B. Optimal operation when Pd, > 0 [80]:Asll f 0, use equatlon (6.6)and sct dPd,lda = 0. This means thatZV,,SI~(O-~+~,) =0IThe corresponding DVR ~njection real power requirement under a,,, controlstrategy is given aspz : ~v,Icos(@)- ~V,,ICOS(Q, - a,, + 6,)6.4. PROBLEM FORMUI.ATIONA powerful PSO based phase advancement compensatlon strategy 1sdeveloped for optimizing the energy storage capacity of the DVR in order to enhancethe voltage restoration property of the device6.4.1. Objective FunctionThe proposed work aims at minimizing the objective function designed toopt~mize the energy injection from the energy storage element of the DVR such as acapacitor or a battery. The mathematical model is changed to the followinggeneralized objective function which is [80] given as,M~nrrnrze Pdvr = Pout - P, (6.25)where Pdvr is the real power supplied by the DVR,respectively.P, and P, are the input power from the source and the load power.

Subject toL,oad voltage advance angle constraint, in whichhad voltage advance angle (a) during each compensation strategy should bewlthin the permissible range6.5. A1,GORITHM OF THE PROPOSED METHOD1. Input the parameters of the system such as three phase voltage magnitudcand angles, supply side voltage angle (6). tlme durat~on of voltage sag,load side power factor angle (0) and number of iterations.2. Specify the lower and upper boundaries of load voltage advance angle (a).3. Init~alize iteration loop, particle position and the particle velocity.4. Calculate the input power flow of each phase (P,.,. Pln2, P,,,,). the totalpower flow (Tp,,) and the power from DVR (P,,,,) for each particle.5. Compare each particles evaluation value. Pdvr, with its pbest. The bestevaluation value among the pbest is denoted as gbest.6. Update the inertia weight W as given by (2.9).7. Modify the velocity V of each particle according to (2.7).If V>Vm" then V = V"""If VcVm'" then V= Vm'"8. Modify the position of each particle according to (2.8). If a particleviolates its position limits in any dimension, set ~ ts position at the properlimit.9. Each particle is evaluated according to its updated position. If theevaluation value of each particle is better than the previous pbest, thecurrent value is set to be pbest. If the best pbest is bmer than gbest, thevalue is set to be gbest.10. If the stopping criteria is satisfied then go to Step 12.1 I. Othmrise, go to Step 4.12. lkparticle that generates the latest gbcst is the optimal value.

6.6 SIMULATION RESULTSThe result of the analysis for the proposed PSO based phase advancementcompensation (PSO-PAC) strategy is illustrated with the following case. Under theproposed PSO-PAC method, DVR uses the power h m the source-side healthyphases to minimize active power supply h m the stored energy source. As anillustration, consider a single-phase sag occuning in a balanced three phase systemwhere the post-sag voltages are (lLO0,lL- 12O0.0.4L - 240" ). respectively. Assumea 2-MVA, 0.85 (lag) power factor load at 22 KV. The load power factor angle can bereadily evaluated and is found to be 31.78". Under presag conditions. each phasesupplies 566.7 KW. During in-phase injection technique. power supplied from thephases is 566.7 KW. 566.7 KW, and 226.6 KW, respectively, while the LWR willsupply the balance power of 340 KW. The source-side input power for the threephases with the phase advancement compensation is 666.6 KW. 666.6 KW. and 266.6KW. respectively. and hence the remaining 100 KW is supplied from the DVRstorage device. The reactive power obtained from the DVR is 210.671 KVAr al theload voltage advance angle a value of 151.7833'. Therefore. from the above results itis clear that with the PAC method. the healthy phases of the source provide moreenergy thus reducing the energy storage burden on the DVR storage device. However.there is a significant increase in reactive power supplied by the IIVR. A reactivepower of 1054 KVAr is supplied by the DVR with the PAC scheme while 210 KVAris supplied with in-phase injection technique.The results of the analysis shown above are considered for the typicalarrangement as shown in Fig. 6.1 where the sensitive load is assumed to have a powerfactor of 0.85 lag and the presag load volIage and current are at 1 p.u. The volIage sagis considered as 60% sag on any one phase of the three phase system. The simulationpameters considered for the above test case are shown in Table 6.1Using the proposed PSO based PAC scheme. it can be seen that the DVR canrestore the voltage sag by reduced real power. Here the real power supplied from

DVR is 99.906 KW for a balanced sag level of 60 %. It is observed that the amount ofreal power obtained in three phases are 666.71 KW. 666.71 KW. 266.68 KW.respectively, and the total power supplied to the load is 1600.1 KW. Thecorresponding value of optimized energy is 832.5511 Wan-hours. The optimum phaseadvance angle obtained using the proposed method is 151.7883'. The reactive powerfrom DVR is 1053.6 KVAr and the line current is 52.48638 A. Hence the proposedPSO based PAC scheme finds the optimum phase advancement angle which is almostclose to that of the angle found by the conventional PAC scheme. This indicates thecorrectness of the proposed method.Table 6.1. Parameters used in PSO method -Single phase sapParametersValuesPopulation Size-.-. - -Vm'" -1801 Acceleration coefficients c, and C) 1 2.0, 2.0

Table 6.2. Comparison of results with PAC and In-Phase injection schemePhase-Advance Proposed PSO based PhaseInjectionCompensation Advance CompensationSchemeScheme (PAC) Scheme(PS0-PAC)1 P ~ I 566.699KW 666.705 KW1 666.71 KW226.679 KW 266.682 KW 266.68 KW1600.1 KW1 EnergyI2811491Watt-h 833.375 Wan-h I--832.5511700.099 WattsI Wan-hFrom the above comparison it is clear that PSO based PAC scheme requiresminimum power for a 60 % voltage sag in any one phase lasting for 30 ms for a inputboltape of 22 KV and saves around 94 W. Based on the ahove analysis. it can he seenthat the amount of storage energy can he reduced. thus resulting in a more economicalrestorer in terms of a more compact design. It is evident that the energy saving fromthe proposed method is significantly better than other conventional methods.

GenerationsFig. 6.3.

i. More reasonable performance measure compared to conventional in-phasevoltage injection and PAC scheme.ii. Energy supplied from the DVR to correct a given voltage sag is reducedwhen compared to conventional techniques.

CHAPTER 7CONCLUSIONThe thesis has investigated the behavior of various particle swarmoptimization algorithms for the management of inherent issues of power systems.Here economic load dispatch. unit commitment. real power loss minimizationensuring voltage stability. power quality improvement using distributed generationand minimization of energy capacity of a dynamic voltage restorer have kenconsidered. The PSO method converges to the global or near global point. irrespectiveof the shape ofthe objective function, like discontinuities. and not smooth functions.The proposed approach is intended for the determination of the global minimasolution and may be implemented as a decision suppon tool for the powermanagement system.The test results of various problems hring out the advantages of the PSOmethod. In the particle swarm optimi~ation method. there is only one population ineach iteration that moves towards the global optimal point. This is unlike classicalevolutionary programming method, which has to deal with two populations. theparents and the children. in each iteration. This makes the PSO methodcomputationally faster. The better computation efficiency and convergence propertyof the proposed PSO approach shows that it can he applied to a wide range of powersystem optimization problems.7.1. RESULTSThe main results of this work are summarized as follows(i) The conventional lambda iterative technique cannot be applied to theeconomic load dispatch problem (ELD) of a power system with combinedcycle cogeneration plants (CCCP) due to the nonsmooth andnondifferentiable nature of fuel cost characteristics. Therefore thecomputationally intelligent algorithms like the proposed approach will be

(ii)(iii)(iv)(v)an efficient way for solving it. The proposed algorithm is demonstratedwith a CCCP system, a 3 bus and a 6 bus system.Economic load dispatch problem has been solved using PSO algorithm.The developed algorithm has been successfully validated with classicaland intelligent techniques of economic load dispatch and hence hasreduced total fuel cost and power loss. The proposed algorithm is appliedin a combined economic emission dispatch environment. The advantage ofproposed algorithm is demonstrated on an IEEE-30 bus system.An efficient and reliable genetic algorithm based particle swarmoptimization technique has been used for economic dispatch pmhlemconsidering prohibited operating zones and ramp rate limits of generatingunits. The integrated GA-PSO combines the conventional PSO frameworkwith the genetic algorithm. The method discourages prematureconvergence to local optimum and also explores and exploits thepromising regions in the search space dTectively. The effectiveness of theproposed algorithm has been tested on a 6. a 15 generating unit systemsand also for a large-scale test system consisting of 40 generating units. Ihesimulation results clearly show that the proposed GA-PSO method can beused as an optimizer providing satisfactory solutions compared to the PSOand GA methods.In the proposed approach. an integrated genetic algorithm method with theparticle swarm optimization technique has been used to solve the unitcommitment problem. The feasibility of the proposed method wasdemonstrated on a 10 unit system and the test results were compared interms of production cost with those obtained by the GA and PSO methods.The real power loss minimization and voltage stability enhancement isanalyzed using a novel PSO based approach for various test systems. Thesw- intelligence technique identifies the best possible voltage profile foran optimum settings of AVR values. OLTC tap positions and for a givennumber of RPCE's to be connected in the system. The improved results ofthe system voltage stability are validated using a line stability index. Theeffectiveness of the algorithm is tested on a Standard-5, IEEE-14, IEEE-30. IEEE-57, IEEE-118 bus systems and for practicaI Indian utilitysystems such a NTPS system and Puducherry bus system. The results

(vi)(vii)obtained using PSO based method is found to provide minimum realpower loss when compared to Newton-Raphson based load flow methodand genetic algorithm approach. The method identifies that PSO algorithmimproves the voltage stability of the power system considerably.The amount of DG capacity to be installed and identification of itsinstalling location is obtained using the swarm intelligence technique. Thetechnique improves the power quality of the system in terms of reducedtransmission losses and improved system voltage profile. Theimprovements in implementing PSO in the existing system are brought outby means of the various power quality indices. The new proposedalgorithm is applied to a standard IEEE -30 bus system and the results arefound to be highly encouraging. The results indicate the reduction in lineloss and a significant improvement in voltage prnfile by using theproposed method when compared with the calculations based on theNewton-Raphson method.DVR is a series compensation device connected to the distrihution lines toovercome voltage sags. The amount of energy injection from the cnergystorage device of the DVR is obtained using various compensationstrategies. The proposed PSO based PAC energy injection technique helpsin optimizing the amount of real power to be injected for a given voltagesag. This in turn reduces the amount of storage capacity required and alsocan improve the ride through capability of the DVR. This methodminimizes the amount of energy injection from the storage device of aDVR to correct a given voltage sag.7.2. FUTURE WORKBased upon the results and discussions of the proposed work, the swarmintelligence based techniques can be applied to test various power systemoperation and control problems by implementing the following extensions:1. Phase shifting transformers are used in controlling the direction and magnitudeof active power flow at inter-tie buses. The various contingencies likely tooccur in a power system have to be studied in order to provide a reliable

power system. Therefore, in a multicomhained economic load dispatchproblem, the optimum power flow can be performed by having phase shifterangle limits and security limits as the constraints. In addition. it may henoticed that valve-point effects in the case of stream turbines cannot hemathematically modeled easily. However it can be considered as a typicalnonsmooth optimization pmblem. Hence hybrid optimization can be adoptedto solve these issues.2. Ramp rate is a limitation for the amount of power generated per hour. Securitylimits and voltage constraints limit at various buses has to he checked toprovide a secured power system. Therefore GA-PSO method can he tested fora unit commitment problem having ramp rate limits. voltage constraints andsecurity limits.3. The proposed method should be used for real pwer loss minimirationensuring voltage stability for various systems having numerous nodes. Thisensures the applicability of method under deregulated environment. Thecontingency analysis can also be performed for simulating various faults. Theperformance analysis can clearly indicate the stability of the system undersuch conditions.4. The optimum DG capacity obtained using the proposed method can alsoincorporate the reduction in environmental pollution due to NO, and C

APPENDICESAPPENDIX - ADATA FOR IEEE-30 BUS TEST SYSTEMThe one line diagram of an IEEE-30 bus system is shown in Fig. A.I. 'TheSystem data is taken from references [I471 (1491. The line data. bus data and loadflow results are given in Tables A.land A.2, respectively. The generator cost andemission coefficients, transformer tap setting, shunt capacitor data are provided inTable A.3, A.4 and A.5, respectively. The B-loss coefficients mauix of the system isgiven in Table .4.6. The data is on I00 MVA base.Fig. A.1. One line diagrnrn

Table A.2. Bus data and Load tlow multsi Bus1 No.--134567Bus VoltageMagnitude(P.U')1.061,0451,0001.0601.0101.0""1,000PhaseAngle(degrees)O.OoO0.0000.0000.0000.0000.0000.000GenerationRealPower(p.u.)0.0000.0000.0000.0000.000ReactivePower(p.u.)1.3848 -0.02790.4 0.50.0000.0000.370.0000.000- .-.0.228 0.109 ---ReactiveLoad PowerLimitsReal ReactiveQlimPower PowerQN,(p.u.) (p.~.) (P-U-) (p.~.)-- -0.000 0.000-.0.217 0.127 4.2 0.60.024 0.012-0.076 0.016----0.942-0.19 4.15 0.6250.000 0.000 - -8 11 - -0I1 112!1131,0101 .0001,0001.0821 .0001,0710.0000.0000.0000.0000.0000.0000.0000.0000.0000.00"0.0000.0000.0000.0000.3730.0000.0000.1620.000-0.30.0000.0580.0000.1 120.30.000-.0.020.0000.075-0.1.5 0.50@I06 0.000 - 0.000 4.15 0.45--0.000 0.062 0.016 - ---4.10-0.40-

Table A.3. Generator cost and Eminsion coe~cients

Table A4. Transformer tap setting dataTable A.5. Shunt capacitor dataTable A.6. Generalized loss coeff~cientsB,", = [0.000014]

APPENDIX - BDATA FOR 6 UNIT TEST SYSTEMThe system contains six thermal units. 26 buses, and 46 trimmission lines1421. The load demand is 1263 MW. The cost coefficients of 6 unit test system arcgiven in Tables B.1. The ramp rate limits of corresponding generating units are givenin Table B.2. The generalized B loss coeficients for the system are shown Table A.3.The system data is on 100 MVA base.Table B.I. Generating Unit Capacity and CoemcienbUnitp,rnln pim.= a((MW) (MW) (SIMW~~)Table B.Z. Ramp Rate Limits and Prohibited Operating Zones

Table B.3. Generalized loss cocflicienta

APPENDIX - CDATA FOR lIUNlT TEST SYSTEMThe 15-unit test system contains 15 thermal units whose characleristics aregiven in Tables C.1, C.2, respectively. The generalized loss cmfiicients an givenTable C.3. The system data is taken from reference 1421.Table C.1. Generating unit with ramp rate limits

Table C.2. Prohibited operating zones of generating unitsUnitProhibited Zones (MW)1Table C.3. Generalized loss coeflicientsI1 (1114 01K112 OM01 4 INX>I 41XX13 4 WL)I 41 OX111 4 IKYII 41 WX13 0 LXXI5 4IKYl1 -llIlKJZ (I IYKU 0 (XU3 4 INKI?0 W112 0iKl13 OlXII1 (I INXNI -11 -0 WW2 11 IXINI II IXXII 4llXN13 41 (XXU -0 IXXM -0 11KXi 11 IIXY IIIXIIII 4IIIXI200(1(11 OWMl 000104 IMII 41NJ13 4lIXXN 4IlYXII IIIKXW U W I H 4IX112 II1X)Il IIIXXXI 4 MI35 0111 11 4lX12Y4 iKKll 4 UXl5 4) W11l 0 1x134 -1, lXW 4 M I 4 ll 1x11 I II(UX0 0 0029 00032 -(I IN11 I -0 (11XI IILXIII 0 INXI1 41 IN126-11 1x11) .IX>~ -11 (mil I w i 7 I~(X~I (1 in114 I (XNI~ II IXII~ n INIU a I~II~ o1m11 -1) Irxc 41 INXC u l x l n 41 (urn4 MII UIXKKI 4 IXIN 41 IXKM o(1114 (I 1~110 I IYYUI -(I INXK a rm5 4 nnlw ~IXIII 41 ( r w n 1~x1: 41 rx,11 elxnl4 MXII 111KK11 -0 (XKII 11 IN11 1 41 (XKl3 41 (KKK1 (I 1X)I 5 11 lXl17 11 IXIlh IIIXXW -11 (XXh 11 11N17 4, 11XXl 4lIXNl? 41 IN118.I, *XI) 11 IXHI o am^ a IX~SI~ 41 IXI? .ii II all1 0016~ o msl IIIXI~V 41 IXI:I 1 IXII~ IIIXYII o IXUK 41 (~1781 4 iKX13 -0 M I 2 -11 WXIX 11 1112'4 -1, IKllll 4 IXXIS 0 (XI15 0 WIX: OOIW 1101 Ih 41Xi21 4, IN125 0 IXU11 -0 IN111 41 (1172-1, lXX13 -11 (KYN -11 W113 I1 11132 -0 (XI13 -0 IXXIX I! IXHIV 0 1Nl7V 0 111 10 OIIZIKI 41 IN127 4) 11114 I,lXNW 4 IXII I 4IlXIWI-0lKKl3 -0WXN -00017 4IINlI I I)IM17 1liX)II 4 IKXIS -0(1123 -00021 41X121 III1141 IIIXXII IIINXU 4IlY138 OlllhX, n NXI? -11 WXI -OIKNX~ 41 IKXN> 4 1 ~ ~ 4 : WXII (I [run 411s% 4 1 ~ 2 11rn14 s OIXXII IIINW atxx)l 1 cwrv 111rn8IIMM O(KH)I .0(1125 (IYUI -1111~12-O~KUI: i ~ ( x ~ m(INNI i IIIYLI~ oixnr, ~IMKU -iiirxit 1,111 111 ~IIIIIII OIKIZXUIXK13 11(Kl1

APPENDIX - DDATA FOR TAIPOWER 40-UNIT SYSTEMThe system data 1s taken from reference [50] whose characteristics am glvenIn Table D.1. The data is on 100 MVA base.Table D.1. Generating units coeff~cients with ramp rate limits

APPENDIX - ELINE VOLTAGE STABILITY INDEXThe important aspect of voltage stability assessment is lo find the distance(MW/MVAR/MVA) to maxlmum loadabtllty point from the present operating point[115]. Line voltage stahil~ty index 1s used to get accurately the proximity of theoperating point to voltage collapse point by the mdcx glven as follows. Let usconsider a single line of an lnterconnccted network. where the lines are connectedthrough a grid network. Any of the lines from that network can be considered to havethe following parameters as shown In Fig. E. 1. Utlliz~ng the concept of power flow Inthe line and analyzing with '11' model representation, the real and reactlve power llowequations in terms of transmiss~on llne constants are formulatedFig. E.1. One line diagram of a typical transmission systemLoads are more often expressed In terms of real (WattsIKW) and reactlve(VArsKVAr) power. Therefore, 11 1s convenient to deal with transmission llneequation in the form of send~ng and receiving end complex power and voltage.Let us treat receiving end voltage as a reference phasor (VR = IVRILO) and letthe sending end voltage lead it by an angle S(Vs = IVslL6). Transrnlssion lines arenormally operated with a balanced 3 phase load. The analysis can be thereforeperformed on per phase basis.The complex power leaving the sending- end and entering the receiving end ofthe transmission line can be expressed on per phase basis [154], as

A transmission llne on a per phase basis, can be regarded as a two-ponnetwork, where in the sendlng end voltage, Vs and current. IS are related to thereceiving end voltage, VK and current. IR through ABCD constants [I541 asReceiving and sending end currents can hc expressed In terms ofreceivlng andscndlng end voltagcs [I 54) as1 =Iv .AVB S B Y(E 3)Let A, B, D the transmission l~ne constants [I 541 he wrltten asA = IAlial. R - IBILPI. 1) - ID1 LUI (Since A =D)Therefore we can write,Substituting for In in equation (E.1) we get,

If equation (E.8) is expressed In real and rlnaglnarv parts, we can write the real andreactive powers at the receiving end [I 541 as.where A La 1 and B LP I are the transnilsston l~ne constantsFor the usual n-model, the transmlsslon l~ne constants may be written as follows(E. 12)If the length of the line is medium, thenZ is the total series impedance of ltne,Y is the total line charging susceptance.If it is a long transmission line, thenZ'Y'A=]+-- 2(E. 13)whereB = Z'

y IS propagation constant and 1 is the length of~ansrn~ss~on llneThe formulae for the receiving end real and reactive powers can he formulated asfollows [I 541These can be rewritten as,(E. I 8)E IEIQ, E~'AISln(p, -a,)= L-Ls~n(p, -8)+--- (E. 19)IBIIBIThen by el~rninatlng 6. by squanng and adding the two equations, we obtainthe locus of PH against QR to be a c~rcle with given values ofA and B and for assumedvalues of ER and jEs/ to be [I 541 as follows.If kth bus is the sending end and m" bus is the receiving end and expanding theabove equation we get as follows,

The above equation should have the real roots for V,,, for the system to bestable. Hence the following condition should bc satisfied[ 1541where, LS, is termed as voltage stability index of the line P, and Q, are the real andreactive power rece~ved at the receiving end rn, ALal and BLP are the transmission11ne constants, Vk and V, are the voltages at the sending end bus k and receiving endbus m.At or near the collapse point, voltage stabil~ty index of one or more lineapproach to unity. This method is used to assess the voltage stability.

APPENDIX - FSTANDARD - 5 BUS SYSTEMThe Standard 5 bus system is shown in Fig. F.I. The System data is takenfrom reference [154]. The line data and bus data are given in Tables F.1. F.2,respectively. The data is on I00 MVA base.Fig. F.I. One line diagrnmTable F.1. Line dnta

Table F.2. Bus Data

APPENDM - GDATA FOR IEEE-14 BUS TEST SYSTEMThe one line diagram of an IEEE-14 bus system is shown in Fig. G.1. TheSystem data is taken from reference [147]. The line data, bus data and load flowresults are given in Tables G.1. and G.2, respectively. The data is on 100 MVA base.C Spchrmxnn Comnprm*orsG Gcnt~atorsheThree Wmdi. Transformer Equivalent-r9Fig. G.1. One line diagram

Table G.1. Line data1 I I I Line lmocdanec I Half Line1ILineTOBusResistance Reactance@-u) (P.u) (P.u.)1 2 0.01938 0.05917 0.02640F~~~BusChargings~~~~~~~~~

Tabk G.2. Bus data and load flow resultsBusNo.123456789-10111213-14Bus VoltageMa'nitude(P.U)1.0601.0451.0101 .OW1.0001.0701.0001.0901.0001.0001.0001.0001.0001.000Generation1:;:(degrees)O.O0OO.OoOO.OOOO.OoOO.OOOO.OoO0.000O.OOO~O.OOOO.OOOO-OoOO.OoOO.OO0RealPower(p.u)2.3240.4000.0000.0000.0000.0000.0000.0000.0000.0000.0000.0000.0000.000Reactivepower(p.~)-0.1690.0000.0000.0000.0000.0000.0000.0000.0000.0000.0000.0000.0000.000LoadRulPower(p.u)0.0000.2170.9420.4780.0760.1 12O.O(H)0.0000.2950.0900.0350.0610.1350.149ReactivePowerReactivepower(p.u)0.0000.1270.1910.0390.0160.075. .0.0000.0000.1660.0580.0180.0160.0580.050Qm,m(Pu)-----0.06~.-.--4.0h-----LimitsQmm,(Ku)-0.500.40-KT'--0.24-----

Table G3. Transformer tap ~niag dataFromBus445ToBus796TapSctting 'Value (p.u.)0.9780.9690.932Table G.4. Shunt capacitor data

APPENDIX - HDATA FOR AN IEEE-57 BUS TEST SYSTEMThe System data is taken h m reference [147]. The line data bus data andload flow results for an 1EEE-57 bus system given in Tables H.1 and H.2.respectively. The transformer tap setting and shunt capacitor data are provided inTable H.3 and H.4, respectively. The data is on 100 MVA base.Table H.1. Line data

Table H.Z. Bus data and load flow results

Table H3. Transformer tap setting dataFromBusToTap SettingValue (P.u.)Bus4 I 18 1 0.97Table H.4. Shunt capacitor data

APPENDIX - IDATA FOR INDIAN UTILITY-NTPS23 BUS SYSTEMThe Indian utility Neyveli Thermal Power Station (NTPS)-23 bus test systemis shown in Figure J.1 .The sites of buses, line data, bus data, me given in Tables 1.1.1.2 and 1.3. respectively. A 100 MVA. 400 KV base is chosen.Fig. 1.1. One line diagram

Tabk 1.1. Sitar and location of different busesI ITV MALAl (230 KV)Thiruvannamalai12CUDDALORE (230 KV)Cuddalore13MDS (400 KV)Madras14151617SLMI (400 KV)SLlvI2 (400 KV)TRY 1 (400 KV)TRY2 (400 KV)Salem 1- . - .Salem2. .-. -- -.'I'richy I.- --- --1-richy218ST1 -ST3 (110KV)Station ~uxillarie;1920D.KURUCH1 ( I I0 KV)VPM 1 &2(110KV)Deva Kuruchi- -Villupuram2 1VDU (1 10 KV)Vadakuthu22P-PDY (1 10 KV)Pondicherry23TVR (I 10 KV)IThiruvarur

Table 13. Line data

Tabk 1.3. Bus data

APPENDIX - JDATA FOR INDIAN UTILITY-PUDUCHERRY- 17 BUS SYSTEMThe Indian utility-Pondicherry-I7 bus test system is shown in Fig. J.1. Theline data bus data, are given in Tables J.l.and J.2.. respectively. A11 0 KV base is chosen.100 MVA.Fig. J.1. One line diagramTable J.1. Line DataHalf Line ChargingL I I I I I

Table J.2. Bus data- - ---- -- -L.oaJ-- - - - --IReactivePower 1

BIBLIOGRAPHYD. P. Kotbari and J. S. Dhilon. Power Svstenr Optmrcutron. New Delhi:Prentice-Hall of Ind~a. 2004.Loi Lei Lai. Inrellrgent Svsrem Applicutrons In Powrr Engmec~rrng, New York:John Wiley & Sons Ltd., 1996.R. C. Bansal. "Optimization methods for electrlc power syslcms:AnOverview," International Journal of Emerg~ng Electnc Power Systems.vol. 2, no. 1, Aniclc 1021,2005.M~chalewicz Zb~gniew, (;enetrc algorrthm + 1)~tu stntcturcv -programs, USA: Springer-Verlag. 1992.Evr~lutionJ. Kennedy and R Eberhart. "Panicle swann opt~m~/atton." In I'roc 199.5IEEE Nc,urol Networkr Conf, vol IV, pp 1042.1 048J. Kennedy and R. Ebcrhan, Swurm Intcllr~r~ncc. LISA. Morgan Kaufn~ann.2001.B. Scott and I. L. Marinho, "Linear programming for power system networksecurity applications," IEEE Trans. Power Apparutus and .Systems, vol. OR,pp. 837-845, MayIJune 1979.0. Alsac, J. Bright, M. Prais, and B. Scott. "Further developments In LP basedoptimal power flow," IEEE Trans. Power Systems, vol. 5, no. 3,pp. 697-71 1, 1990.G. T. Heydt and W. M. Grady, "Optimal VAR siting using linear load flowfonnulatiorr," IEEE Trans. Power Apparatus and Systems, vol. 5. PAS-102,pp. 1214-1222, 1983.G. D. Inisari, L. M. Kimball, K. A. Clements, A. Bagchi, and P. W. Davis," Economic dispatch with network and ramping constrants via interior pointmethods, " IEEE Trans. Power Systems, vol. 13, no. I, pp. 236-242, 1998.

L. S. Vargas. V. H. Quintana, and V. Vannell~. "A tutonal descnpt~on of anIntenor point method and its applications to saunty constra~r~cd economlcdispatch, " IEEE Trans. Power Svstems, vol. 8, no. 3, pp. 1315- 1324. 1903.S. Granv~lle. "Optimal reactive dispatch through ~ntenor pant methods." IEFETruns. Power $stems. vol. 9, no. 1. pp. 136-144. 1994Jcsus Rlquelme Santos. Alic~a Troncoso Lora. Antonlo Gomet Ekposlto. andJose LUIS Martinez Ramos, " Flnding improved local mln~ma of' powcr systcnlopt~mizat~on problems by interior-point methods," IEEE Truns PowrrSystems, vol. 18. no. 1, pp. 238-244, Feb. 2003.M. Christoforidis, M. Aganagic. B. Awobamise. S. Tong. and A F. Rah~m~."Long-tedMid-term resource optimization of a hydro-dom~nant powcrsystem ustng interior polnt methods," IEEE Trans I'ower S\:rtems, vol I I.no. I, pp. 287-294, 1986.J. C. 0. Mello, A. C. G. Melo, and S. Ciranv~lle, "Simultaneous capah~l~tyassessment by combining intenor point mcthod and monte carlo slmulat~on. "IEEE Trans Power Svstems, vol. 12. no. 2, pp. 736.747. 1997K. Ponnambalam, V. H. Quintana. and A. Vannell~, "A fast algorithm forpower system optimization using interior point method." IEEE Trans. PowerSvstems, vol. 7, no. 2, pp. 892-899, 1992.V. R. Sherkat and Y. Ikura, "Experience with interior point optimizationsoftware for a fuel plannlng appl~catlon," IEEE Trans. Power Systems. vol.9.no.2, pp.833-840, 1994.J. A. Momoh, S. X. Guo, E. C. Ogbuoriri, and R. Adapa. "The quadraticinterior point method solving power system optimization problems," IEEETrans. Power Systems, vol. 9, no. 3, pp. 1327-1 336, 1994.J. Nanda, D. P. Kothari, and S. C. Srivastava, "A new optimal power dispatchalgorithm using fletcher's QP method, " IEE Proc.- Generation. Transmissionand Distribution, vol. 136, no. 3, pp. 153-161, 1989.G. Opoku, "Optimal power system VAR planning" IEEE Trans. PowerSystem, vol. 5, no. 1.w. 53-59.1990.

A. M. Chebbo and M. R. Irving. "Combined actlve and rcxtnc pou,crdispatch Part 1. problem formulation and solutron algonthm. " IEE I'rw -Generatron. Trunsmrss~on and Drstrrhw~on, vol. 142. no. 4, pp 393-4(X).1995.A. M. Chebbo, M. R. Irving. and N. H. Dandach~, "Comhlned actlve andreact~ve power dispatch Part 2: test results," IEE Proc-(;

J. F. Dapao and H. M. Merill. "Optimal generator scheduling w~th Integerprogramming, " IEEE Trans Power Apparatus and L'$:vrem.v. vol. 93, no 7. pp1537-1545, 1975Y. H. Song, Modern oprrrnlirrrron rechnrques rn power sysrenas. London, CIKKluwer Academic, 1999F. C. Lu and Y Y Hsu. "React~ve powerJvoltage control In a d~str~hurronsubstation using dynamic programm~ng." IEE Proc - (;enerurron Tran.smr.rsronand Drstrihut~on. vol 142, no 6. pp. 639-645. 1995.J. Parten, "A simplified modified dynam~c prob~amming algonthm for s171nglocation and feeder re~nforcements." IEEE Trans. Powrr Delrven, vol 5.no. I, pp. 277-283, 1990.W. J. liobbs, G Hernlon, S. Warner, and Ci. I3 Shehle. "An enhanceddynamic approach for unit comm~tment." IEEE Trans Powrr Swrems. vol. 3,no. 3,pp. 1201-1205, 1988.K. Y. Lee and M. A. El-Sharkawl, "Modem heurist~c optlmlzatlon techniqueswith applications to power systems," IEEE Power Engtneer~ng Socrer)(02TP160). 2002.P. N. Suganthan, "Particle swam optlm~zer with neighborhood operator. " InProc. 1999 Evolurronary (bmpululron Congress. pp. 1958- 1962. 1999.Carlos A. Coello, Gregorio Toscano Pulido and Maximino Saluar Lechuga,"Handling multiple objectives with particle swarm optimization," IEEE Trans.Evolut~onaty Compuralron, vol. 8, no. 3, pp. 256-279, June 2004.P. Angeline, "Using selection to improve particle swam optimi7at1on." inProc. 1998 IEEE Evolu~lonary Computation Con$, pp. 84-89.Y. Shi and R. Eberhart, "Parameter selection in particle swarm optimization",in Proc. 1998 Evolufionary Programming, Con$. pp. 591 -600.

A. I. Wood and B. F. Wollenberg, Power Gwerutron Openrrron und Conrrol.New York: John Wiley & Sons, 1996.Zwe-Lee Gaing, "Part~cle swarm optimization to solving the economlcdispatch considenng the generator constraints," IEEE Trans. Power Swrems.vol. 18, no. 3. pp. 1 187- 1 195. Aug. 2003.D. N. Jeyakumar. T. layabarath~, and T. Raghunathan. "Particle swannoptimization for vanous types of economic dispatch problems." InrerntrrrrmirlJournal of Elecrrrcal Power und Energy Svsrems. vol. 28. no 1, pp 36-42. Jan2006.J. W. Talaq, M. E. El-Haway, and F-El Hanvay. "A summary oSenvironmentalleconom~c d~spatch algonthms." IEEE Trrms. F'ow,c*r S,:vrc.nrs.no. 3, pp. 1508-1 516, 1994K. P. Wong and J. Yurev~ch, "Evolut~onary programming based algor~thm forenv~ronmentally - constra~ned economic dispatch." IEEE Trims I'on~crSvstems. vol. 13, no. 2, pp. 301 -306. May 1998P. Venkatesh, R. Gnanadass, and Nardyana Prasad Pady, "Companson andapplicat~on of evolutionary programming techniques to comb~ned economicemission dispatch w~th I~ne flow constra~nts," IEEE Trans Power Sv.vremr.vol. 18, no. 2, pp. 688-697, Feb. 2003.M. A. Abido, "Multiobject~ve evolutionary algorithms for electnc powerdispatch problem," IEEE Trans. Evolurronury C'ompururion, vol. 10, no. 3, pp.315-329. June 2006.D. C. Walters and G. B. Sheble, "Genetic algorithm solution of economlcdispatch with valve point loading," IEEE Trans. Power Sysrems. vol. 8,no. 3, pp. 1325-1332, Aug. 1993.K. P. Wong and Y. W. Won& "Genetic and genetidsimulated-annealingapproaches to economic dispatch." IEE Proc.- Generation. Transmissron andDishibution, pt. C, vol. 141, no. 5, pp. 507-513, Sep.1994.

P. H. Chen and H. C. Chang, "Large scale economic dlspatch by genetlcalgorithm," IEEE Trans Power Svstems, vol. 10. no. 4. pp. 19 19- 1926.Nov. 1995.C. C. Fung. S. Y. Chow, and K. P. Wong, "Solving the economic dlspatchproblem with an integrated parallel genetic algorithm", in Proc.2001) IEEEPower System techno lo^ ('ottf, no.3. pp.1257-1262.El-Gallad, M. El-Hawary, A. A. Sallam. and A. Kalas. "Swarm ~nrell~genccfor hybnd cost dispatch problem", In Proc. 2001 ('anadtan Elecrrrc.ulComputer Engrneerrng ConJ. vol. 2. pp. 753-757El-Gallad, M. El-Hawary, A. Sallam, and A. Kalas, "Particle swarm optlmlrerfor constrained economlc dispatch w~th prohibited operating zones." In I'roc2002 Canadtan Electrtcal Cbmptrter Engmeermg Conf, vol. I, pp. 78-81.L. L. Lai, T. Y. Nieh, D. Vujatovic. Y. N. Ma, Y. P. Lu, Y. W. Yang, and H.Braun, "Part~cle swarm optlmlzatlon for economic dlspatch of units w~th nonsmoothinput-output characterist~c functions," in Proc. 2005 Infcllrgc~nt S,af~.tnAppltcafrons lo Power S)jsfems. pp. 499-503.T. A. A. Victoire and A. E. Jcyakumar, "Reserve constralncd dynam~cdispatch of units w~th valve-point effects," IEEE Truns Power ,Sv.srems.vol. 20, no. 3, pp. 1273-1282. Aug. 2005.Tim T. Maifeld and Gerald B. Sheble, " Genet~c-based unit commltmentalgorithm," IEEE Trans. Power Sysfems. vol. 1 1, no. 3, pp. 1359- 1370,Aug. 1996.S. A. Kazarlis, A. G. Bakirtz~s, and V. Petridis, "A Generic algorithm solut~onto the unit commltment problem", IEEE Trans. Power Sy.~tems, vol. I I,no. I, pp. 83-92, Feb.1996.K. S. Swarp and S. Yamashiro, "Unit commltment solution methodologyusing genetic algorithm," IEEE Trans. Power Systems, vol. 17, pp.87-91, Feb.2002.

Zwe-Lee Gaing, "Discrete par~lcle swarm optimization algonthm for unitcommitment." in Proc 2003 IEEE Power Er~grneerrng SrK-lety C~enerulMeeting, vol. 1, pp. 418-424.B. Zhao. C. X. Guo, B. R. Bai, and Y. J. Cao, "An improved parttcle swarmoptimization algorithm for urn1 commitment," lnrernarronul Journul ofElectrical Power and Energv mr.ms, vol. 28, issue I, pp. 482-490. Jan.2006T. 0. Ting, M. V. C. Rao. and C L. h, "A novel approach for unltcommitment problem via an effectlvc hybrid partlcle swarm opt~mizat~on."IEEE Truns. Power Svsrems. vol. 2 1. no. I, pp. 41 1-41 8. Fcb.2006.Ahmed Yousuf Saber, Tomonohu Senjyu. Atsushl Yona, Naomltsu llrasak~.and Toshihisa Funabashi, "Fuz~y unlt commitment solut~on-A novel twofoldsimulated annealing approach." Elcctrrc Power S,:srcms Rc,st.rrrch. vol. 77.pp. 1699-1712.2007.T. A. A. Vlcloire and A. E. Jeyakumar. "Un~t conlmltment by a tabu-searchbasedhybrid-opt~mizatlon technique," IEE f'roc - (;encrutron. Tron.srnr.svrorrand Dumhutlon, vol. 152. no .4, pp 503-574. Jul. 2005.H. Yoshida, K. Kawata, Y. Fukuyama, S. Takayama, and Y. Nakanlshl, "Aparticle swarm optimlzatlon for reactlve power and voltage controlconsidering voltage securlty assessment," IEEE Truns. Powrr S\arcm.s,vol. 15, no. 4, pp. 1232-1239, Nov. 2WXl.V. Miranda and N. Fonseca, "EPSO-evolutionary particle swarm optlmi~.at~on.a new algorithm w~th applicat~ons In power systems", in Pr~r. 2002 IEf/I/Pf:.YTransmission D~strrhutron ConfErencc Exhrhitron. Asru-Puc$c. vol. 2,pp. 745-750.H. Mantawy and M. S. Al-Ghamdi, "A new reactrve power optimizationalgorithm", in Proc. 2003 IEEE Power Tech. ConJ. vol. 4, pp. 6-1 1Ahmed A. A. Esmin, Germano Lambert-Torres, and Antbnio C. Zambronl deSowa,"A hybrid particle swarm optimization applied to loss powerrni~mization", IEEE Trans. Power Systems, vol. 20, no. 2, pp. 859-866, May2005.

M. Tripathy and S. Mishra, "Bacteria foraging-based solution to opt~mize bothreal power loss and voltage stability Iim~t," IEEE Truns. Power Svsrems.vol. 22, no.1. pp. 240-248, Feb. 2007.Yair Malachi and Sigrnond Singer, "A genct~c algorithm for the correctivecontrol of voltage and reactive power," IEEE Truns Power S~fstcms, vol. 21.no. I, pp. 295-300, Feb.2006.Amgad A. EL-Dib. Hosam K. M. Yousscf. M M EL-Metwally, wd 7..Osman, "Optimum VAR sizlng and allocat~on uslng part~cle swarmoptimization." Ncctrrc Powcr S~srerrrs Rc.~carc~h vol. 77, pp. 005-972. 2(H)7.R Ramakumar and P. Chiradeja, "Distributed generation and renewableenergy systems", tn Pror 2002 3P Intcrsorrc!~ Enrrm Cbn,,crsronEngmcenng Con/. pp 716-724Pathomthat Chiradeja and R. Ramakumar. "An approach to quant~ly thetechnical benefits of distributed generation," IEEE Trans. Enerm (bnvcr.sron.vol. 19, no. 4, pp. 764-773, Dcc. 2004G. Joos. B. T. Ooi, D. McCill~s, F. D. Galiana. and K. Marccau, "The potentla1of distributed generation to providc anc~llary services," in Proc 2000 IEEEPower Engmeer~ng Socrcty Srrmmer Mcetrn~. vol. 3, pp.1762-1707.R. Rarnakumar and P. Chirade~a, "A probabil~sttc approach to the analysls ofvoltage profile ~mprovement with distributed wind electric generation, " inProc 2001 Frontrers ojPowcr Con/, pp. XI1 1 - XI1 10.G. Celli and F. Pilo, "Optimal dlstnbuted generation allocation in MVdistribution networks," in Proc. 2001 IEEE Power Engmeerrng Society PowerIndustry Computer Applrratrons Cont, pp. 8 1-86.J. A. Greatbanks, D. H. Popovic, M. Begovic, A. Pregelj, and T. C. Green,"On optimization for security and reliability of power systems with distributedgeneration," in Proc. 2003 IEEE PowerTech ConJ

S S. Chot, B. H. Li and D. M. Vilathgamuwa. "Dynamlc volcagc mstorationw~th minimum energy injection," IEEE Trans Power SIIIP~S. vol. IS. no. 1,pp. 51-57, Fcb. 2000.Jr. Alexander Domijan. Alejandro Montenegro. Alben J F. Ken. and KennethE. Mattem. "Custom Power Devices: An lnteractlon Study. " IEEE Trun.sPower Svstem.~, vol. 20. no. 2. pp. I I I I - 1 I 18. MayM. H. Haque. "Voltage Sag Correction hy Dynam~c Voltage U~cstorer withmrnimum Power Injection." IEEE Pou.rr Engr~rc.c.rrn,g Kcr.tcw, pp 5(,-58.2001.D. Mahlnda V~lathgamuwa A. A. D. Uan~~th Perera. and S S Cho~. "Voltagesag compensatlon with energy opt~mized dynannc voltage restorer," IEEETruns Power Delivrn, vol. 18. no. 3. pp 928-936. Jul. 2003D. Mahinda Vllathgamuwa, H. M. Wijekoon. and S S. C'ho~. "A Noveltechn~que to compensate voltage sags In mult~l~ne d~stnbut~on system l'heinterline dynamic voltage restorer,'' IEEE Truns Indrcsrrrol EIccrronrcs,vo1.53, no. 5, pp.1603-1611, Oct. 2006.P. Venkatesh, P. S. Kannan. and M. Sudhakaran,"Appl~cat~on ofcomputational intelligence to economlc load dispatch," IE (Inrlru) Journul oflnstrtutron of Engrneers, vol. 8 1, pp. 39-43. Scp 20CKI.M. Sudhakaran, S. M. R. Slochanal, "Integrating gcnetlc algorithms and tabusearch for emission and economic dispatch problems." IE (Indru) Journul 4Imrrtution of Engrneers, vol. 86, pp. 39-43. June 2005.R.C. Eberhan and Y. Shi, "Particle swarm optlml~ation. developmcnts,applications and resources", in Proc.2001 Evolurron Cbmpulutron Congress.pp. 81-86.C. A. Ro*Sepulveda and B. J. Pavez-Lazo, "A solution to the optimal powerflow using simulated annealing, " in Proc. 2001 IEEE Porto Power Tech ConfS. 0. Orero, M. R. Irving, "A Genetic algorithm for generator scheduling inpower systems," International Journal of Electrrcul Power and EnermSystem, vol. 18, no. I, pp. 19-26, Feb. 1996.

[87] D. B. Fogel. "Evolut~onary Computation: A new Tr;ms;act~on." IEEETrans Evolurronary Compuratron, vol. 1, no. I. pp. 1-2. Apr. 1997.[88] S. Baskar and P. Subburaj. "Performance of hybr~d real coded Cicnet~calgorithms," IE (Indra) Journal of lnsr~tutron of Engmc,t*rs. vol 81, pp 25-31.Nov. 2000.D. B. Fogcl, Evolurronary Comnpufarron. Toward u hicw Pkrlosopliy r!/Muchrne Intelligence. 2nd. ed.. Plscataway. NJ. IEEE Press. 2000R. C. Eberhart and Y. Shl, "Companson bctween gnetlc algor~thms andpart~cle swarm opt~mization." in Pr(x.1998 IEEE E~r>lutron ('ompututronConf, pp. 611-616.Chao-Lung Chlang, "Improved genetic algorithm for power cconomlcdispatch of units with valve polnt effects and mulr~ple fuels." IEEE Truns.Power Sv.~rems, vo1.20. no. 4, pp. 1690-1699, Novcmbcr 2005.P. S. Kulkami, A. G. Kothari, and D. P. Kothar~, "Con~blncd cconom~cemiss~on dispatch using improved back propagation ncural network."lnternatlonal Journal of El~ctrrcul Maclrlne.~ and Power. S,Ttems, vol 28.pp. 3 1-43,2000.Gerald B.Sheble and George N. Fahd, "Un~t comm~tment l~tcraturesynopsis,"IEEE Trans. Power Systems, vol. 9, no. I, pp. 128-135. Feb. 1994.John J. Shaw, "A direct method for security constrained unlt comm~tment,"IEEE Trans. Power Svstems, vol. 10, no. 3, pp. 1329-1 339, Aug. 1995.S. K Tong, S. M Shahidehpour, and 2. Ouyang, "A heuristic short-term unitcommitment," IEEE Trans. Power Systems, vol. 6, no. 3, pp. 1210-1 216, Aug.1991.Jong-Hwan Kim and Hyun Myung, "Evolutionary programming techniquesfor constrained optimization problems," IEEE Trans. EvolutronaryComputation, vol. 1, no. 2, pp. 129-140, Jul. 1997.

[97] Chuan-Ping Cheng. C'hth-Wen 1.1~. and Chun-Chang Liu. "CJmt comnutmentby lagrangian relaxation and genettc algonthms." IEEE Trans. Power .S~stewts.vol. 15, no. 2, pp. 707-714, May 2000.[98] A. H Mantawy. Youssef L. Abdel-Mag~d, and Shokn Z. Sel~m. "lntcgratlngGA and SA for the unit commltment problem." IEEE Trans Pou,er J.~:\.tt.rns.vol. 14, no. 3, pp. 829-836, Augl999.[99] N. P. Padhy, "Unit commitment-a bibliogrdphical survey." IEEE Tran.~ Pow,crSy.~tems, vol. 19, no. 2. pp. 1 196- 1205. May 2004.[I001 S. 0. Orero and M. R. Irvlng, "A Genettc algorithm for generator schedulingin power systems," lntrrnatronal Journal of Elecrrical Pon8c.r ant1 Encr~,Systems, vol. 18, no. 1. pp. 19-26, Feb. 1996.[loll G. B. Sheble, T. T. Matfeld, K Hr~tt~g, and G. Fahd, "Untt commltment bygenetic algorithm with penalty methods and a comparison of lagrangtan searchand genetic algorithm - economic dispatch example. " Intrrr~arronrrl Journrrlof Power and Enerp s.v.vterns. vol. 18. no. 6, pp. 339-346. 1096.[I021 Carolina M. Affonso. Luiz C. P. da Sllva. Flavio G. M. Lima. and SccundtnoSoares, "MW and MVAR management on supply and demand s~de (ormeeting voltage stability margln criter~a," IEEE Trans Power .Stst~,rn.s,voL 19, no. 3, pp. 1538-1545. Aug. 2004.[I031 Wei Yan, Juan Yu, Dav~d C. Yu, and Kalu Bhattarai. "A new reactive powerflow model in rectangular form and its solution by predictor corrector primaldual interior point method," IEEE Trans. P0wc.r Systems. vol 21. no. I, pp.61-67, Feb. 2006.[I041 Saikat Chakrabarti and Benjamin Jeyasurya, "An enhanced radial bastsfunction network for voltage stability monitonng considenng multtplecontingencies, " Electr~c Power System Research, vol. 77. no. 7, pp. 780-787,2007.[I051 Ivan Smon, Gregor Verbic, and Ferdinand Gubina "Local voltage-stabilityindex using Tellegen's theorem", IEEE Trans. Power Systems, vol. 2 1, no. 3,pp. 1267-1275, Aug. 2006.(1061 T. Esaka, Y. Kataoka, and T. Ohtaka. "Voltage stability preventive controlusing a new voltage stability index," in Proc. 2004 Power system TechndogvConJ, pp. 344-349.

[I071 Costas Voumas and M~chael Karystianos. " Load tap changers in Emergencyand Preventive voltage stab~lity control," IEEE Trcms Powcr S\sr~~nts. vol 19.no. I, pp. 492-498, Feb. 2004.[lo81 LUIS A. LI. Zarate, Carlos A. Castro. Jose LUIS Martlnej. Ramos. and EstherRomero Ramos, " Fast computation of voltage stability security niarg~ns uslngnon-l~near prograrnmlng techniques. " IEEE Trans Po~r-r S):\rcm.s. vol. ?I.no. I, pp. 19-27, Feb. 2006.[I091 Antonio J. Conejo. Federico Milano, and Raqucl C;arcia-Hcrtrind." Congestion management ensurlng voltage stability, " Trllns IJon,crSystems, vol. 21, no. I, pp. 357-364, Feb.2006.(1101 G. Krost and G. A. Bakare, "A genctlc algorithm based approach forImprovement in voltage profile and real power loss rnlnlmlmllon. " In I'roc1999 EIecrr~c Powcr Englneerrng Conf. pp. 153.[I1 I] Saikat Chakrabartl and Benjam~n Jeyasurya, "Multi Contingency voltagestabil~ty mon~toring of a power system uslng an adapt~vc radlal hasls funct~onnetwork," Internu~ronul Journal of Elcctrtcul f'ower u~id Encrx~, .S~~.stemv,vol. 30, no. I, pp. 1-7, Jan. 2008.[I121 J. Y. Wen, Q. H. Wu, D. R. Turner. S. J. Cheng, and J. C~tch, "Opt~malcoordinated voltage control for power system voltage stability." IEEE Truns.Power Svstems, vol. 19, no. 2, pp. 11 15-1 122, May 2004.[113] Pablo A. Ruiz and Peter W. Sauer, " Voltage and reactive power estlrnatlon forcontingency analys~s ustng sensitivities," IEEE Truns. Power Systems, vol. 22.no. 2, pp. 639-647, May 2007.[I141 Yasushi Koyama, Tetsuo Sasaki, Satoru Ihara, and Elizabeth R. Prat~co,"Voltage Collapse Scenario Search," in Proc. 2OU2 IEEE Power SystemTechnology Conf, pp. 344-348.

[I 151 S. C. Choube. L D Arya, and N. Datar. "Voltage Collapse Pred~ctlon Rasedon Line Voltage Stabtltty Index," IE (India) Journal Insrtr~crlon ofEngineers, vol. 82. pp. 107-112. Scp. 2001.[I161 M. Suzuk~, S. Wada, M. Sato, T. Asano. and Y. Kudo. "Ncwly dcvclopzdvoltage secunty momtonng system." IEEE Trans. Powcr Svsrcms. vol 7.no. 3, pp. 965-972, Aug. 1992.[I 171 Anh~t Sode-Some. Nadarajah Mtthulananthan, and Kwang S. I.ec."Amaximum loadtng margln method for statlc voltage stah~ltty In powrrsystems," IEEE Trans. Power Svsrems. vol. 21, no. 2. pp. 965.972. May 2000[I181 Arth~t Sode-Yome. Nadarajah Mtthulananthan, and Kwang Y Lee." Econom~c generatton dtrect~on for power system statlc voltage stab~l~ty. " InProc 2006 IEEE Power Engrncerlng Socletv (;cnercll Mccfr~tg. Vol 8. No 22.PP 1-7[I 191 P. Ch~radeja and R. Ramakumar, "A review of distr~buted generat~on andstorage," In Proc. I998 Fro~lrlers of Powcr Con/crenrc, pp. Vlll 1 - 1 1[I201 P. Chir~dejand R. Ramakumar. "Benefits of dtstnbutcd generatton - a stmplecase study," in Pror I999 Fronrrers cfPower (bn/erence. pp. X 1-9[I211 N. Vaitheeswaran and R. Balasubramanian. "Stochast~c model for optimalselection of DDG by mont carlo simulation." In Proc. 2006 PowerElectronics. Drives and Energv ,Svslem Con/. pp. 1-5[I221 R. C. Dugan and S. K. Pnce, " Issues for d~stributed generations In Ihe US," InProc. ZOO2 IEEE Power Engzneering Societ.~ W~nrer Meermg, vol. 1,pp. 121-126.[I231 R. C. Dugan and T. E. Mcdermont, "Distributed generation, " IEEE Indusrr~ulApplication Magazine. pp. 19-25, March /April 2002.[I241 T. Gray, "Wind gea competitrve in the US," Solar Today, vol. 12, no. 2.pp. 18-21, MarcWApril 1998.

[I251 W. G. Scott, "M~cm-turb~ne generators for distributed systems." IEtXIndustrial Applrcatlons Maga:rne. pp. 57-62. MayIJune 1998.[I261 T. Raissi, A. Banerjee. and K. G. Scheinkopf, " Current technology of fuel cellsystems," in Proc. 1997 Intcrsorretv Energy Conwrsion Engmecring ('onf .pp. 1953-1957.(1271 S. Rahman. " Fuel cell as a d~stnbuted generation technology." In Proc 2(K)1IEEE Power Engmeerrrrz Socreh, Summer Meetmg. vol I, pp 55 1 552[I281 T. Hoff and D. S. Shugar. "The value of gr~d-support photovolta~cs Inreducing distribut~on system losses," IEEE Trr~ns. Dwgv C'or11,r~rsron. vol10, pp. 569-576, Sep.1995.[I291 Y. G. Hegazy. M. M. A. Salama and A. Y. chi khan^. "Adequacy assessmentof d~stributed gcneratlon systems uslng monte carlo s~mulation." IEEE 7nm.sPower Svstems, vol. 18, no. 1, pp.48-52. Feh.2003.[I301 J. P. Koen Macken. Math H J. Bollen, and Ronn~e J. M. Belmans, "M~t~gat~onof voltage dips through distnbuted generatton systems," IEEE Tr~ns Industn,Applicarlons, vol. 40, no. 6, pp. 1686-1693, Nov./Dec. 2004.[I311 Aleksandar Pregelj, Miroslav Begovic and Ajeet Rohatg~, "Quant~tattvctechniques for analysis of large data sets in renewable d~strihuted generat~on,"IEEE Trans Power Systems. vol. 19. no. 3, pp. 1277-1285. Aug.2004.[I321 Caisheng Wang and M. Hashem Nehriw, "Analytical approaches for optimalplacement of distributed generation sources in power systems," /EEL TransPowerSystems, vol. 19, no. 4, pp. 2068-2076, Nov. 2004.[I331 Victor H. Mhdez Quezada. Juan Rivier Abbad, and Tomh Gomez SanRomh, "Assessment of energy distribution losses for increasing penetrationofdistributed generation," IEEE Trans. Power Systems, vol. 21, no. 2,pp. 533-540, May 2006.

[I341 P. Chiradeja, "Benefit of Distributed Generation: A Line Loss Rcduct~onAnalysis, " in Proc 2005 IEEE/PES liunsmission and Distrihurton Conj undExhibitton: Asia and Pacrfic. pp. 1-5.[I351 T. Jauch. A. Kara, M. Rahman~, and D. Western~ann. " Power quallry etisuredby dynam~c voltage correction. " ARB RFIJ, VOI. 4, 1998.[I361 N. A. Samra, C. Nefl, A. Sundaram. and W. Malcolm. "The d~str~hut~o~~system dynamic voltage restorer and its applications at industrial fac~ht~eswith sensitive loads." In Proc IYYS Power Conversion Intel1 Morton f'owrtr@ally Con/:[I371 Alexander Kara, Davld Atnhof Peter Dahler, and Horst Gmnlng. " Powcrsupply quality improvement wlth a dynamic voltage restorer." in Prur 1998Apphed Power Electronrcs Con/. Vol. 2, pp. 080-993.[I381 S. W. Middlekauff and Jr. E. K. Collinges. "System and customer Impact:consideration for series custom power devices," /EKE Truns. f'ower D~~bvrn,,vol. 13, pp. 278-282, Jan. 1998.[I391 Poh Chiang Loh, D Mahinda V~lathgamuwa. Seng Khai Tangand H~an 1.1hLong, " Multilevel dynamic voltage restorer." in Proc. 2004 Power .~vstemTechnology ConJ, pp. 1673-1678.[I401 John Godsk Nielsen and Frede Rlaabjerg, "Control strategies for dynam~cvoltage restorer compensating voltage sags with phase jump," In Proroc.. 2001IEEE Applied Power Electronics ConJ, vol. 2. pp. 1267-1 272.[I411 Hyosung Kim, "Minimal energy control for a dynamic voltage restorer," inProc.2002 IEEE Power Conversion ConJ, pp. 428-433.[I421 Chris Fitzer, Mike Barnes, and Peter Green. " Voltage sag detect~on techniquefor a dynamic voltage restorer," IEEE Trans. Industry Appltcatrons, vol. 40,no. 1, pp. 203-212, Jan./Feb. 2004.

[I431 F. Jurado and Perez Hidalgo. "Neural network control for dynamrc voltagerestorer." in Proc. 2002 IEEE lndustr~ul Electronics Socleiy Con/. vol. I,pp. 61 5-620.[I441 T. Wundcrlin, P. Dahlcr. D. AmhnC and H. Grunrng. "Power supply qual~tyimprovement wrth a dynam~c voltage restorer (DVR)," In ProcIYYh'EMPD '98 Conj. pp. 5 18-525[I451 H. Woodley, L. Morgan, and A. Sundaram. "Experience with an Inverterbaseddynamic voltage restorer." IEEE Truns Pon,r.r I)clrwn,. vol. 14, pp1181.1 186, Jul. 1999.[I461 1'. K. Nagsarkar and M S. Sukh~ja. Ponx.r sy.ytcm unuhni.~, New Dclhl:Oxford University Press. 2007.[I471 Ray. D. Zirnmerman, Carlos E.Munllo-Sanchez, and tlcq~ang (Davrd) Gan."Matpower-Power system simulatron package," versron 3 0. Power systenisEngineering Research Center (PSERC'), School of Elcctncal Engrneer~ng.Cornell University, Ithaca, Ncw York, 2005.[I481 0. Alsac and B. Stott, "Optimal load flow w~th steady state sccunty." IEEETrans. Power Apparatus untl S)srem.s, vol. PAS-03, no. 3, pp.745-751. 1974.[I491 David 1. Sun, Bruce Ashley, Bnan Brewer, Art Hughes. and Willram F.Tinney, "Optimal power flow by newton approach." IEEE Truns. PowerApparatus and Systems, pp. 2864-2875, Oct. 1984.[I501 M. Anitha, S. Subramanian, and R. Gnanadass, "FDR PSO-based transrentstability constrained optimal power flow solution for deregulated powerindustry," Electr~c Power Components and System, vol. 35, no. I I,pp. 1219-1232. Nov. 2007.[I511 Takashi Mitani, Yuji Mishima, Taiji Satoh, and Koichi Nara. "Optimalgeneration scheduling under competitive environment," in Proc. 2006 IEEEon Systems, Man and Cybernetics ConJ. pp. 1843- 1848.

[I521 S. K. Joshi and K. N. Patel, "Real tlme economlc dlspalch." In Proc 000IEEE Power System Technologv Cbnf, vol. 3. pp 1263-1768.[153] K. Nithiyananthan, V. Ramachandran. "Component model for multi-areapower systems on-line economic load d~spalch," Inrernurronul Journul ofEmerging Electrrc Power Systems, vol. 1. no. 2, article 101 1.2004.[I541 S. Scnthil Kumar, "Accurate method of comput~r~g l~ne voltage stability hascdindicator of voltage collapse In a power system. " M.E Thcs~s, Departrncnt ofElechical Engineering. Annamalai University, Dec. 2002.

LIST OF PUBLICATIONSlnternational Journals1. "Particle Swarm 0ptimii.ation Based Optimal Power Flow Solution forCombined Economic Emission Dispatch Problem". lnternational Journal ofElectrical Systems. Vo1.3, No.1, pp.13-25. March 2007.2. "GA and PSO Culled Hyhrid Technique for Economic Dispatch Problem withProhibited Operating zone". An lnternational Applied Physics endEngineering Journal. Journal of Zbejiang University Scienee A. Springer.Vo1.8. No.6, pp. 896-903. June 2007.3. "Swarm Intelligence for Voltage-Var Compensation Pn)hlem", lnternationalJournal of Computational Intelligence: Theow and Practice, Vol.2. No.'.pp.135-140.2007.4. "Swarm Intelligence for Voltage Stability Analysis Cnnsidering Voltage-VARCompensation Problem", Journal of Electrical Engineering, Elektrika, (Inpress: Manuscript no. JE2007-43).5. "Integrated GA-PSO based Unit Commitment", International Journal ofEngineering Research and Industrial Applications, Vol.1. No.4, July 2008.(In press: Manuscript no. EN 191).National Maenzine1. "Swarm Intelligence Based Real Power Loss Minimiation", Electrical India,Vo1.48, No.4, pp.118-123, April 2008.lnternational Conferences I National Confemnces1. "Investigations on Combined Economic Emission Dispatch Problems byEvolutionary Computing Techniques", Proceedings of lnternationalConference on Computer Applications in Electrical Engineering (CERA2005). Indian Institute of Technology, Roorkec. India. pp. 120-123.September 29th -I st October. 2005.2. .'Particle Swarm Optimization Based Optimal Power Flow Solution forCombined Economic Emission Dispatch Problem", Prdings of 14"

National Power System Confmnce (NPSC 2006). Indian Institute ofTechnology. Roorkee. India, pp 44, C3- 4, December 27th -29th. 2006.3. "Application of Particle Swarm Optimization for Economic Load DispatchProblems", Proceedings of 14m International Confcrence on IntelligentSystem Applications to Power Systems (ISAP 2007). Taiwan. pp.668-674.November 4th - 8th. 2007.4. "Swarm Intelligence based Optimization of Distributed Generation Capacityfor Power Quality Improvement", Proceedings of lnternational Confcrenceon Advances in Energy Research, IIT. Mumbai, India pp.513-519. kcember12th -14th. 2007.5. "Particle Swarm Optimization bawd Energy Optimized Dynamic VoltagcRestorer", Proceedings of International Conference on Power SystemAnalysis, Control and Optimization (PSACO 2008). Andhra liniversit).Visakhapatnam, India. pp.764-771. March 13th - 15th. 2008.Communicated:1. "Swarm lntelligcnce based Optimization of L)istributod Generation Capacityfor Power Quality Improvement" IET Renewable I'ower Generation -(Manuscript no.: KPG-2008-0041).2. "Integrated Genetic Algorithm and I'article Swarm Optimiration for solvingllnit Commitment Problem". IEEE PCS, International conference on Powersystem technology. Powercon 2008 -(Manuscript no.: 0047).3 "Swarm Intelligence based Energy Optimized Dynamic Voltage Restorer".Journal of Electrical Engineering. Romania - (Manuscript no.: 88).

VITAEP.Ajay-D-Vimal Raj born in Karaikal. India in 1976. He received his R.Edegree in Electrical and Electmnics Engineering from Madras University. Chennai inthe year 1998 and subsequently obtained his M.E degree in Power Systems fmmAnnamalai University, Chidamharam in 1999. He is presently working as a Lecturerin the Department of Electrical and Electronics Engineering, Pondicherry tngineeringCollege. Pondicherry and hw nearly eight years of teaching experience. Hr is a lifemember of the Institution of Engineer (India). Indian Society for Technical Education(ISTE) and Society of Power Engineer (India). He has to his credit over twenty livepapers in nationallinternational conferences and journals. His area of interest includesApplicution of Intelli~enr 7Pchnique.s to I'o~,c,r S~:stc~m.s opcJrution und controlproblems undpou,er syvrcvn oprirnizurion trc.hniyui~.\