A Differential Protection Method for Intelligent Substations Based on Multi-Scale Feature Extraction and Temporal Modeling

With the continuous advancement of smart grid and digital substation technologies, traditional differential protection methods have gradually exposed limitations such as delayed response and high misjudgment rates when dealing with complex operating conditions and nonlinear transient disturbances. To enhance the robustness and intelligence of transformer differential protection, this paper proposes an intelligent differential protection identification method that integrates Neural Network-based Principal Component Analysis (NNPCA), Convolutional Neural Network (CNN), and Lightweight Gated Recurrent Unit (LGRU). Firstly, NNPCA is used to perform nonlinear dimensionality reduction and decorrelation processing on sampled differential current signals, thereby enhancing feature expression capabilities. Then, a multi-scale CNN module is constructed to extract local spatial features of disturbances, and an LGRU network is introduced to model the temporal evolution of differential current, enabling dynamic and accurate identification of internal faults and non-fault disturbances. Simulation results demonstrate that the proposed method outperforms traditional protection strategies in terms of accuracy, response speed, and anti-disturbance performance, showing strong practical value and engineering applicability.