Application of semi-supervised fuzzy kernel clustering algorithm in recognizing transformer winding's pressed state

Abstract

This paper applies fuzzy clustering algorithm to recognize the transformer winding's pressed state based on transformer's vibration signal. We propose a new semi-supervised fuzzy kernel clustering algorithm (SFKC) based on some modifications for the fuzzy clustering methods. The first modification is that the new algorithm uses prior knowledge to guide the clustering process. Second, it uses kernel function to map the samples to high dimensional feature space for clustering. Third, dynamic weight of the feature is carried out considering the different effects of sample features. The accuracy and reliability of the proposed algorithm are verified by the standard test data set. Then the algorithm is applied to recognize transformer winding's pressed state. According to the vibration characteristics of the transformer, we construct a sample set incorporating multi-sensors and multi-features for clustering. After clustering, we use the clustering centers and feature weights to recognize new unlabeled sample. The results show that the method is feasible.