In order to obtain more abundant discharge characteristic information and improve the efficiency of par-tial discharge diagnosis,a deep learning fusion method of partial discharge in oil-paper insulation based on multi-spectral information is proposed.Firstly,based on the micro-optical sensor and the spectral distribution of partial discharge,a multi-spectral synchronous detection platform for partial discharge is constructed,and the multi-spec-tral data of four discharge types are obtained through experiments.Then the convolution neural network model is constructed,and the partial discharge data of different spectral sections are used as the input of different channels of the model.The effective information in the multispectral signal is extracted by channel level fusion,and the par-tial discharge types of oil-paper insulation are accurately identified.The results show that the multi-spectral infor-mation of different discharge types can be used as an effective feature of pattern recognition;by the introduction of multi-spectral information,the recognition accuracy of the proposed method can reach more than 98%,which is significantly improved compared with that of only using pulse current signals;compared with statistical characteris-tic parameter analysis and deep neural network,and the proposed method has better effect on multi-spectral infor-mation fusion and higher recognition accuracy.
维普
万方数据