SVM-based classification methodology for overhead distribution fault events

This paper proposes the application of support vector machines (SVM) to classify overhead distribution faults according to their general root causes; they are faults due to animal contacts, tree contacts, and lightning-induced events. The SVM method uses unique features buried in voltage and/or current waveforms. Seven unique features based on time and electrical quantities are presented. The performance of support vector machines with different kernels is compared to that of a rule-based classification method. The training and classification results demonstrate that SVM-based approach performs better than the rule-based approach. For instance, SVM-based approach correctly classifies 119 out of 148 collected voltage events, whereas rule-based approach 88 out of them. Likewise, a good generalization performance of the SVM-based approach is demonstrated during the training process carried out. However, the drawback of such a classifier based on SVM, and other blackbox methods, is due to the difficulties to interpret decision criteria.