Improved Particle Swarm Optimization-based Support Vector Machine for Fault Diagnostic of Arrester

Arrester plays an important role in protecting equipment used in power system against overvoltage phenomena. Thus, diagnosic of arrester condition is attracting considerable attention from the researchers to ensure the safety and effectiveness of electric power distribution systems. In this paper, a newly perturbed particle swarm optimization (P-PSO) is proposed to classify the fault of surge arrester by adjusting the parameters of the support vector machine (SVM). The proposed method is employed on an actual dataset which consists of 1600 patterns with three features each (the total leakage current its resistive component and the third harmonic of resistive leakage current). The experimential ressults show that the proposed P-PSO is ability to not only accurate diagnostic arrester's faults, but also avoid the local minima. To demonstrate the superiority of the proposed method in identifying the conditions of the arrester, the obtained results are compared to those used by the other variants of PSO such as classical PSO (CPSO), time-varying acceleration coefficients PSO (TPSO) and constriction factor PSO (KPSO).