Transformer condition assessment based on combined deep neural network

Dissolved gas analysis (DGA) occupies an extremely important position in transformer condition assessment, and many conventional methods have been proposed based on dissolved gas in oil analysis. In this paper, a combined deep neural network (CDNN) is proposed to combine the characteristics of dissolved gas in oil and conventional methods for transformer condition assessment. First, the sample data are normalized according to the characteristics of the conventional method parameters. Then, the normalized parameters are used as input parameters of the deep neural network. Multiple deep neural network models are built separately. Next, the prediction results of multiple deep neural network models are weighted according to the accuracy of different models. Finally, the one with the largest weight is selected as the final prediction result of the combined deep neural network. In this paper, we demonstrate the necessity of data normalization through data statistics and demonstration. The comparison between setting up the normal state data and not considering the normal state data proves that the normal state is easy to misclassify with other states when predicting, which leads to the decrease of the prediction accuracy. By comparing with the composition method of this paper and the classification method commonly used in MATLAB software, it is confirmed that the method in this paper combines the advantages of other methods and corrects the prediction results, thus having higher accuracy.