Feature Extraction of Partial Discharge Signal Based on Local Mean Decomposition and Multi-scale Singular Spectrum Entropy

Feature extraction of partial discharge signal is a key step in pattern recognition and fault diagnosis of insulation defects in power equipment. The singular-spectrum entropy (SSE) theory can study the complexity and irregularity of partial discharge signals but cannot fully reflect the inherent nonlinear characteristics of signals. On this basis, the covariance matrix of SSE is replaced by the Hankel matrix of partial discharge signals, and the local mean decomposition (LMD) theory is introduced to realize the multi-scale method. Then, a multi-scale Hankel singular spectrum entropy (MULTI-HSSE) method for partial discharge signals is proposed. Through the analysis of the simulated partial discharge signals, the entropy eigenvectors extracted by this method can effectively improve the noise suppression ability and enhance the robustness of phase space reconstruction parameters. Finally, three typical partial discharge defects are designed in an outdoor substation environment. The UHF signal entropy eigenvectors using this method are achieved and the RBF neural network is used to classify the defects. Through the experiments, the high recognition accuracy is verified, showing the validity and applicability of the proposed method.