Joint Prediction of Goaf Temperature and CO Concentration Using Multi-Source Monitoring Feature Fusion and CA-WOA-Optimized Models

Variations in goaf temperature and CO concentration during low-temperature oxidation are jointly affected by residual-coal oxidation, air-leakage oxygen supply, gas generation, migration, dilution, and ventilation disturbance, resulting in nonlinear multi-source responses. This study investigated a joint prediction framework based on multi-source monitoring feature fusion and covariance-adaptive whale optimization algorithm (CA-WOA)-optimized models. Daily monitoring data from goaf pipes, the working face, the upper corner, and the return-air side were integrated. Pearson correlation analysis and random forest feature-importance ranking were used to construct top-k feature subsets. CA-WOA incorporates rank-weighted elite-center reconstruction, covariance-adaptive direction learning, WOA random-search injection, and geometric step-size decay, and was used to optimize RF, XGBoost, LightGBM, CatBoost, LSSVM, and TABM under a unified dual-output fitness function. CA-WOA improved all six models, with a maximum mean R2 increase of 0.076 and a maximum fitness reduction of 22.8%. CA-WOA–TABM achieved the best fivefold performance, with a mean R2 of 0.924 and F=0.273, while its out-of-fold R2 values for temperature and CO concentration were 0.928 and 0.931, respectively. On the additional 1202 working-face dataset, TABM achieved a later-period test RCO2 of 0.766 under chronological splitting, and CA-WOA–TABM obtained a fivefold mean RCO2 of 0.948. SHAP and PDP analyses identified GoafPipe_CH4, GoafPipe_CO2, GoafPipe_C2H6, and GoafPipe_O2 as the main predictive variables.