Partial Discharge Pattern Recognition of XLPE Cable Based on Stacked Sparse Noise Reduction Autoencoder

The traditional machine learning algorithms have already been implemented on the partial discharge pattern recognition of cross-linked polyethylene (XLPE) power cables. However, the slow convergence speed and low recognition accuracy limit its practical engineering applications. To solve the drawbacks, this paper presents an XLPE cable partial discharge pattern recognition method based on a stacked sparse noise reduction autoencoder. This proposed method determines the hyperparameters by category weights and adds Gaussian white noise to the original input. The autoencoder can fully extract effective features, and then obtains an effective deep feature extraction model. The Softmax classifier is employed to deliver the diagnosis results. The analysis results of the case study show that compared with the partial discharge pattern recognition methods such as decision tree, BP neural network and SVM, the proposed method can further improve the recognition accuracy by adding Gaussian white noise and determining the optimal number of hidden layers.