Application of Fuzzy Entropy to Improve Feature Selection for Defect Recognition Using Support Vector Machine in High Voltage Cable Joints

This study presents a method for defect-recognition in high voltage cable joints based on partial discharge (PD). This recognition involves three major systematic procedures. In the first procedure, the PD patterns are produced by two different laboratory models representing two types of defects in a high voltage cable. The PD data are collected from a set of experiments in the PD tests with six high voltage cable joints, including prefabricated artificial defects. The second part involves feature selection by employing a fuzzy entropy algorithm by which the entropy value of each defect is computed. Using this fuzzy entropy algorithm, the features that have the most useful characteristics for distinguishing the defects in cable joints are found. In the third part, the selected features are used for testing and training the support vector machine (SVM) model, and the accuracy testing rates are calculated in order to obtain optimal results. The SVM model in this study achieves a higher accuracy rate of 96% for classification with the proposed feature-selection-based fuzzy entropy algorithm.