Abstract: Based on the actual situation of frequent faults in electrical secondary equipment during the operation of hydropower stations, a fault warning system integrating artificial intelligence algorithms was constructed. By introducing wavelet analysis, EMD decomposition, and CNN deep feature extraction mechanisms, multi-scale modeling of equipment status data is carried out. Combined with LSTM and attention mechanism, a temporal prediction model is constructed to achieve accurate early warning of hardware, software, and external interference faults. The research results show that in actual operating environments, the average warning accuracy of the system reaches 93.3%, with an early warning time of over 60 min, and has good real-time performance, robustness, and engineering adaptability.