Partial discharge development stage division based on multi-classifier fusion

Abstract

The oil-paper insulation air gap discharge experiment was carried out by the step-up method, and the 29-dimensional discharge characteristic parameters were extracted. The oil-paper insulation air gap discharge experiment was carried out by the step-up method, and the 29-dimensional discharge characteristic parameters were extracted. According to K-means clustering, the partial discharge development stage is divided into four stages: initial discharge stage, discharge development stage, discharge stability stage and near breakdown stage. The three basic classifiers of random forest, support vector machine and back propagation neural network are trained by the credibility-based adaptive weighted voting algorithm to train a new fusion classifier. Compared with the traditional classification fusion device, the proposed classifier fusion model can effectively improve the accuracy of the identification of the discharge development stage, up to 87.5% or more.