Fault Early Warning for Secondary Cabinets Based on Sensor Data and Machine Learning

Secondary cabinets are core control units in power systems, and their operating status directly affects the safety and stability of power grids. To address the problems of experience-dependent and inefficient traditional methods, this paper proposes a fault early warning scheme that integrates multi-source sensor data and machine learning. Sensors for temperature, humidity, vibration, and other variables are used to collect operating data. Preprocessing and feature engineering are then performed to extract key information. The XGBoost model is used for fault identification, and dynamic thresholds are used for early warning. Field validation shows that the proposed scheme achieves an early warning accuracy of 95.2% and an average warning lead time of 4.2 hours. The fault detection rate is 87.3% higher than that of traditional methods, which effectively reduces operation and maintenance costs and fault losses.