A new data mining approach to dissolved gas analysis of oil-insulated power apparatus

Abstract

This paper proposes genetic algorithm tuned wavelet networks (GAWNs) for data mining of dissolved-gas-analysis (DGA) records and incipient fault detection of oil-insulated power transformers. The genetic algorithm-based (GA) optimization process automatically tunes the parameters of wavelet networks: translation and dilation of the wavelet nodes, and the weighting values of the weighting nodes. The GAWNs can identify the complex relations between the dissolved gas content of transformer oil and corresponding fault types. The proposed GAWNs have been tested on the Taipower Company's diagnostic records, using four diagnosis criteria, and compared with artificial neural networks (ANNs) and conventional methods. Experimental results demonstrate that the GAWNs have remarkable diagnosis accuracy and require far less learning time than ANNs for different diagnosis criteria.