Relay Protection Fault Detection Method Based on Time-Frequency Feature Collaborative Enhancement and Multi-Kernel SVM Graded Warning

Fault signals of relay protection feature strong nonstationarity, and traditional algorithms fail to extract time-frequency features and grade fault risks. This paper proposes a hierarchical early warning scheme integrating synergistically enhanced time-frequency features and multi-kernel SVM. Adaptive complementary ensemble empirical mode decomposition denoises and screens modes; time-domain transient and wavelet packet frequency-domain features are fused into joint vectors for a three-level fault classification multi-kernel SVM model. Tests show its response curve matches actual conditions, with 9.2 ms total time and 97.0% average grading accuracy, outperforming conventional algorithms comprehensively.