Risk-oriented interpretable prediction of regional dust exposure in a fully mechanized heading face under long-forcing and short-exhaust ventilation

To improve near-field dust exposure risk identification and ventilation-parameter regulation in fully mechanized heading faces, this study proposes an interpretable prediction method for dust exposure in key zones. Based on 1296 computational fluid dynamics (CFD) cases under long-forcing and short-exhaust ventilation, regional average concentrations were extracted for the driver zone, return-side breathing zone, heading hazardous zone, and floor-sensitive zone. The over-threshold fraction in the heading hazardous zone was further used to characterize the spatial expansion of high-concentration dust near the source. Source-strength, particle-size, ventilation, and interaction features were constructed to establish machine-learning mappings between operating parameters and regional exposure indicators. Shapley additive explanations (SHAP) analysis was then used to identify dominant factors and their physical response mechanisms. The results show that the model achieved satisfactory predictive performance for the average concentrations in the floor-sensitive zone, return-side breathing zone, and heading hazardous zone, with test-set R² values of 0.93, 0.88, and 0.84 and mean absolute percentage error (MAPE) values of 7.61%, 8.82%, and 6.72%, respectively. The over-threshold fraction in the heading hazardous zone achieved an R² of 0.80, indicating its potential for assisting the identification of high-concentration source-zone expansion. SHAP results indicate that source-strength parameters dominated all five core indicators, particle-size parameters mainly affected particle transport-deposition redistribution, and the pressure-exhaust ratio modified dust migration paths and risk transfer through ventilation parameters and interaction features. Using the average concentration in the heading hazardous zone as the criterion, the low-, medium-, and high-risk classification achieved an accuracy of 0.862 and a macro-F1 score of 0.851. The proposed framework provides an interpretable basis for regional exposure assessment, risk screening, and ventilation regulation.