Fault evolution prediction model and intelligent diagnosis algorithm driven by power equipment operation data

Abstract

Aiming at the challenge of fault evolution prediction and intelligent diagnosis in power equipment operation and maintenance, this paper proposes a joint model driven by multi-source operation data. By fusing heterogeneous data such as electric quantity, oil chromatography and temperature, an improved health index (HI) is constructed to quantify the degree of equipment degradation. Adopting a bidirectional long short-term memory network (BiLSTM) combined with attention mechanism to capture temporal degradation trends and accurately predict future health status and remaining useful life (RUL); A multi-task learning framework is designed to optimize fault evolution prediction and type recognition, and the diagnosis accuracy is improved by using degradation trajectory information. The experimental results show that the mean square error (MSE) of the model in HI prediction is 0.0087, the coefficient of determination (R2) is 0.954, and the fault diagnosis accuracy is 94.3%, which is significantly better than traditional single task models and benchmark methods. Ablation experiments verify the key roles of attention mechanism, multi-source data fusion and multi-task combination in improving performance. This study provides an efficient and reliable solution for predictive maintenance of power equipment.