Research on fault prediction method of power equipment based on multimodal feature fusion and machine learning

Power equipment fault prediction is crucial to the reliability of power grid and operation and maintenance costs, but the difficulty of modeling multi-dimensional characteristic dynamic correlation and the lack of model interpretability restrict its application. In this paper, an ensemble framework that integrates dynamic binning, association rules and Bayesian networks is proposed: optimizing feature discretization based on industry standards, mining high-confidence rules to construct interaction features, generating causal probability features (BN_Prob) through Bayesian networks, and integrating XGBoost and SMOTE balance data. Experiments show that the accuracy of the proposed method is 94%, which is 10% higher than that of the independent model, and the scheme has both high accuracy and interpretability, and can be extended to intelligent manufacturing and other fields, promoting the evolution of predictive maintenance to causal decision-making.