Multiple Classifiers Based Information Fusion for Power Transformer Fault Diagnosis

Abstract

Power transformer fault diagnosis is quiet significant for power grid reliability. Traditional diagnosis methods mostly rely on a single classifier to diagnose transformer faults, by analyzing the dissolved gas. However, due to the complexity and variability of the dissolved gas, these methods with a single classifier may fail to exploit the dissolved gas information to diagnose faults effectively. To tackle this issue, in this paper, a multiple classifiers based information fusion (MCIF) method is proposed for power transformer fault diagnosis. Specifically, this method consists of two main steps. First, multiple classifiers are employed to generate initial diagnosis results. Different classifiers can capture the dissolved gas information from different aspects, providing complemental statistics information for fault diagnosis. Second, we employ a decision-level information fusion method to combine initial diagnosis results, so as to produce a fused diagnosis result. Experiments verities the proposed method can achieve a competitive performance, compared with some classical and outstanding methods including extreme learning machine (ELM), support vector machine (SVM), and probabilistic neural network (PNN).