A Hybrid Fault Early-Warning Method Based on Improved Bees Algorithm-Optimized Categorical Boosting and Kernel Density Estimation

In the context of intelligent manufacturing, equipment fault early-warning technology has become a critical support for ensuring the continuity and safety of industrial production. However, with the increasing complexity of modern industrial equipment structures and the growing coupling of operational states, traditional fault warning models face significant challenges in feature recognition accuracy and adaptability. To address these issues, this study proposes a hybrid fault early-warning framework that integrates an improved bees algorithm (IBA) with a categorical boosting (CatBoost) model and kernel density estimation (KDE). The proposed framework first develops the IBA by integrating Latin Hypercube Sampling, a multi-perturbation neighborhood search strategy, and a dynamic scout bee adjustment strategy, which effectively overcomes the conventional bees algorithm (BA)’s tendency to fall into local optima. The IBA is then employed to achieve global optimization of CatBoost’s key hyperparameters. The optimized CatBoost model is subsequently used to predict equipment operational data. Finally, the KDE method is applied to the prediction residuals to determine fault thresholds. An empirical study on a deflection fault in the valve position sensor connecting rod of the mineral oil system in a gas compressor station shows that the proposed method can issue early-warning signals two hours in advance and outperforms existing advanced algorithms in key indicators such as root mean square error (RMSE), coefficient of determination (R2) and mean absolute percentage error (MAPE). Furthermore, ablation experiments verify the effectiveness of the strategies in IBA and their contribution to CatBoost hyperparameter optimization. The proposed method significantly improves the accuracy and reliability of fault prediction in complex industrial environments.