Partial discharge pattern recognition of current transformers using an ENN

This paper proposes an extension-neural-network (ENN)-based recognition method to identify the partial-discharge (PD) patterns of high-voltage current transformers (HVCTs). First, a commercial PD detector is used to measure the three-dimensional (3D) PD patterns of cast-resin HVCTs, then three data preprocessing schemes that extract relevant features from the raw 3-D PD patterns are presented for the proposed ENN-based classifier. The ENN proposed in the author's recent paper citation combines the extension theory with a neural-network architecture. It uses extension distance instead of using Euclidean distance (ED) to measure similarities between tested data and cluster centers; it can implement supervised learning and give shorter learning times and simpler structures than traditional neural networks. Moreover, the ENN has the advantages of high accuracy and noise tolerance, which are useful in recognizing the PD patterns of electrical apparatus. To demonstrate the effectiveness of the proposed method, comparative studies with a multilayer multilayer perceptron (MLP) are conducted on 150 sets of field-test PD patterns of HVCTs with rather encouraging results.