Partial Discharge Pattern Recognition Based on Stacked Denoising Autoencoder Network: Computer and AI Applications in Power Industry

Since partial discharge (PD) detection on power site can be easily affected by various disturbances, the detection data are always polluted with noise, thus are hardly to extract the obvious features for traditional recognition methods. This paper gives a new method which is based on Stacked Denoising Autoencoder Network (SDAE) for partial discharge. An SDAE model is established, which actively add noise on the typical defect partial discharge data from experiment in the model training process. The model can extract deep feature of partial discharge data with noise, and output the recognition result with Softmax classifier. A contrast experiment is designed on PD data substations. The experimental results show that the presented method has a higher recognition rate in dealing with noising partial discharge data.