Fault Diagnosis Method of Power Transformers Based on Wavelet Networks and Dissolved Gas Analysis

The wavelet network (WN) is the novel and efficient model of nonlinear signal processing developed in last years. This paper presented the type 1 and type 2 WN mathematical models for power transformer faults diagnosis. In addition, the WNs based on an adaptive algorithm was proposed which inherited learning ability of artificial neural network and localization characteristics of wavelet transform, so it has good convergence property and robustness. After the two types of WNs based on different mother wavelet functions were developed from the 250 groups of training and recognizing gases-in-oil samples by fuzzy preprocessing, comparison and analysis about training process and simulation results were carried out. A lot of diagnosis examples show that the diagnosing performance of the proposed two types of WNs approaches are suitable for faults diagnosis of power transformers and prevail that of traditional BP neural network.