Wavelet Networks in Power Transformers Diagnosis Using Dissolved Gas Analysis

Wavelet networks (WNs) are an efficient model of nonlinear signal processing developed in recent years. This paper presents a comparative study of WN efficiency for the detection of incipient faults of power transformers. After 700 groups of training and testing gases-in-oil samples are processed by fuzzy technology, we compare and analyze the network training process and simulation results of five WNs which include two types of WNs with two different activation functions and evolving WN. A lot of diagnostic examples show that the diagnostic accuracy and efficiency of the proposed five WN approaches prevail those of the conventional back-propagation neural-network method and are suitable for faults diagnosis of power transformers, especially with the evolving WN achieving superior performance.