1.State Key Laboratory of Disaster Prevention & Reduction for Power Grid, Changsha University of Science & Technology, Changsha 410114, China;2.State Grid Yiyang Power Supply Company, Yiyang 413000, China;3.College of Electrical and Information Engineering, Hunan University, Changsha 410082, China;4.Zhongshan Power Supply Bureau, Guangdong Power Grid Co., Ltd., Zhongshan 528400, China

The increasing integration of power electronic devices in distribution networks has exacerbated harmonic currents, challenging the reliability of traditional transformer differential protection based harmonic carrent restraint. Additionally, the single-characteristic identification methods are affected by the distributed generator (DG) types and closing angles, and cannot accurately distinguish fault currents and inrush currents in various scenarios. In order to improve the identification accuracy of inrush current, this paper proposes a multi-perspective time-frequency analysis method that comprehensively integrates time-domain, frequency-domain, and time-frequency-domain characteristics. The Bayes algorithm is used to optimize the classification parameters of extreme gradient boosting (XGBoost), improve the generalization ability of the model, and achieve accurate identification of fault current and inrush current with different capacities and types of DG connected. The SHapley Additive exPlanation (SHAP) value analysis method is used to reveal the contribution of each characteristic value in the identification model. Based on simulation and on-site measurement data, the proposed inrush current identification method is validated, and the accuracy of identification for sample data is close to 100%.