Design and Implementation of Facts Controller for Power Quality Improvement

Abstract

ABSTRACT newlineUnified Power Quality Conditioner (UPQC) is one of the Flexible AC newlineTransmission System (FACTS) devices which is used to mitigate various types of newlinevoltage distortion in electric load. The configuration of UPQC is the combination of newlineshunt active power filter and series active power filter which shared with dc-link newlinecapacitor. When Power Quality (PQ) problem occurs, the voltage across the dc-link newlinecapacitor is varied from the rated voltage. Numerous intelligent controllers are used newlinefor regulating dc-link voltage. Among them, Adaptive Neuro-Fuzzy Interference newlineSystem (ANFIS) is one of the recent controllers. To improve the effectiveness of newlineUPQC control, Genetic Algorithm (GA) based Neuro-Fuzzy Controller (NFC) is newlinedeveloped. Here, GA is used for optimizing the membership function of fuzzy newlinesystem. But, the traditional GA is unable to ensure the stable optimization response newlineall the time. In addition, the dissimilarity of shortest and longest optimization time is newlinehigher than with conservative grade methods by the reason of generation of single newlinechromosome. Consequently, an Adaptive Genetic Algorithm (AGA) is designed to newlinegenerate sub-chromosomes for the entire genetic operation. Thus, the subchromosome newlinegenes are migrated until a best fitness value is achieved. As a result, an newlineoptimal solution is obtained possibly and accurately than the traditional GA. The newlineoutput of Adaptive GA is used for developing Neuro-Fuzzy interference system newlinewhich produces an optimal dc-link regulation voltage. But, the computational newlinedifficulty of Adaptive GA is elevated than classical GA, as it takes more time to newlinereach the solution. Thus, Cuckoo Search (CS) based NFC is designed for developing newlinethe performance of UPQC to compensate these problems. CS is used for optimizing newlinethe output of neural network so the classification output of the neural network is newlineenhanced. The inputs of the networks are error and change of error voltage of the newlinenonlinear load which is calculated by comparing with the reference signal. From the newlinenewly designed CS-NFC cont