Partial Discharge Pattern Recognition Algorithm Based on Sparse Self - coding and Extreme Learning Machine

The traditional partial discharge pattern recognition method has low recognition accuracy or long training time. In this paper, a new PD pattern recognition method is proposed. The PRPS map of the PD signal is extracted as the input data. The feature extraction and dimension reduction of the PRPS map are realized based on the Sparse Autoencoder (SAE) to obtain low-dimensional feature space that can express the original data highly. The Extreme Learning Machine (ELM) network is used as a classifier to realize the classification of PD. The experimental results show that the proposed algorithm not only has a high accuracy of pattern recognition but also has a fast training speed. Key words: PD; SAE; ELM; pattern recognition