Research on building a fault feature model of current sampling unit based on short circuit impulse analysis

Under the “dual carbon” strategy, power system refined scheduling demands higher measurement accuracy from smart energy meters. This paper addresses the challenges in fault diagnosis and model adaptability of current sampling units under short-circuit impacts. We systematically investigate fault feature modeling based on short-circuit impulse analysis. The nonlinear damage mechanisms of short-circuit impacts on current transformers, manganese copper shunts, and analog-to-digital converter modules are analyzed, establishing correlations between impact intensity, signal distortion, and parameter drift. A hierarchical preprocessing method combining time-frequency analysis and 1D-DCNN enables dynamic fault feature extraction, with key features selected via recursive feature elimination. Multi-intensity short-circuit tests are designed, and high-fidelity 3D modeling with finite element simulation reveals thermal distribution and fault evolution. The optimized BP neural network and LS-SVM classification model achieves over 92% recognition accuracy. This research provides technical support for fault warning, lifespan assessment, and impact-resistant design of current sampling units, ensuring meter accuracy under extreme grid conditions.