Assessment of Different Multiclass SVM Strategies for Fault Classification in a PV System

Abstract

Fault detection and diagnosis is an imperative choice for the long life of a PV system. The conventional protective devices fail to detect possible faults, owing to the non-linear nature of the voltage-current characteristics of the PV system impelling the need for a better technique. A novel approach to classify PV faults, making decision boundaries using SVM, propelled by dimension reduction using PCA is demonstrated for classifying different fault classes. Faults considered for classification are short-circuit fault in any module, inverse bypass diode fault, shunted bypass diode fault, and shadowing effect in a module. SVMs are binary classifiers and involve meticulous effort for extending the theory to more than two classes. The paper highlights the efficiency and runtime complexities of the various multiclass SVM techniques like One versus One, One versus All, Decision Directed Acyclic Graph, and Adaptive Directed Acyclic Graph. Methods are compared for the results of different `training: testing data samples' (60:40, 70:30, 80:20) using synthetic PV data points from PVLIB toolbox.