Simulation Research on Fault Location in Small Current Grounding Systems of Distribution Networks Based on Feature Fusion

Abstract

This paper focuses on the research of fault location under different operating conditions and noise interference. Due to the limitations of single features, this paper proposes a multi-feature fusion fault location method based on Extreme Learning Machine (ELM) to improve the effectiveness and accuracy of single-phase grounding fault location in distribution networks. This method extracts three features - sample entropy, high-frequency energy proportion, and Hilbert phase difference from the current signal and fuses them into a data feature set. ELM is used as the classifier, and the training and testing results are obtained. The training and testing are conducted under different operating conditions to verify the generalization ability of the model. The results show that the fusion of these three features is significantly superior to a single feature, and it still performs well in complex operating conditions, which helps to improve the accuracy of fault location.