Research on Comprehensive Identification and Classification Technology for Latent Faults in High- Voltage Switchgear Based on Machine Learning

With the development of ultra-high voltage projects in China, the scale of power grids has been expanding. High-voltage switchgear is widely utilized in various substations and the power industry, gradually becoming a critical component of the power system. In response to the issue of latent faults in high-voltage switchgear, this paper proposes a pulse current detection method based on machine learning to achieve detection and classification of latent faults in high-voltage switchgear. The Hilbert-Huang transform is employed along with the theory of energy entropy to extract discharge characteristic parameters, which serve as input vectors for the pattern recognition network. The discharge types are recognized using a Probabilistic Neural Network (PNN), and optimization algorithms are applied to enhance the performance of the PNN. The recognition accuracy before and after optimization is compared. The results indicate that although this method requires a longer recognition time, the recognition accuracy is relatively high, reaching 98.7%. This approach plays a significant role in achieving comprehensive identification and classification of latent faults in high-voltage switchgear.