Bayesian Optimized Convolution Neural Network Architecture for Partial Discharge Signature Recognition

The detection, measurement, and classification of Partial Discharges (PD) is an important method to determine the quality of insulation systems used in High-Voltage power apparatus and cables. The PD detector patterns contain the characteristics of an insulating system malfunction. Through PD, the problem is found by taking relevant information out of it. In this proposed work, Convolutional Neural Networks (CNN) extract valuable features from a 3-D Phase-resolved Partial Discharge pattern, where their edges are detected using a canny edge detected approach. CNN performance is improved by optimizing the algorithm hyperparameters. Finally, the proposed technique is compared to cutting-edge CNN algorithms from the literature. According to the findings, the proposed method can be utilized to identify possible challenges, attaining an accuracy of 97.3%.