PD Pattern Recognition in Transformers Based on Grey-scale Images and Affinity Propagation Algorithm

Abstract

Partial discharge (PD) pattern recognition plays an important role in determining insulation defects and understanding insulation condition of transformers. In this paper, four PD models are set up in laboratory and pulse current method is used to measure the amplitude of apparent charge, the power frequency phase of PD pulses and the frequency of PD pulses. There are 19 feature parameters which include fractal features, moment features, and textural features are extracted form grey-scale images of phase resolved partial discharge (PRPD) patterns. In order to reduce the computational complexity of the classifier, principal component analysis (PCA) is used to reduce the dimensions of feature parameters and five new feature parameters which explain 93.50% of total variance are obtained. The kernel function and shared nearest neighbors (SNN) are used to improve affinity propagation (AP) algorithm. A classifier based on improved AP with particle swarm optimization (PSO) is established for PD pattern recognition in transformers. Based on these new feature parameters, the PD patterns are recognized by AP classifier, back propagation neural networks (BPNN) and least squares support vector machine (LSSVM). Recognition rate of AP classifier is 85%