Machine learning-based regulation of nitrogen/phosphorus-containing explosion-inhibitors for inhibiting methane/coal dust hybrid explosions

This study explores the inhibitory effects of nitrogen/phosphorus structure-controlled composite materials on methane/coal dust explosions. By integrating experimental data and chemical descriptors, it utilizes the Extreme Gradient Boosting (XGBoost) model, optimized through Particle Swarm Optimization (PSO), to enhance prediction accuracy. The model achieved R² improvements of 1.11 %, 0.7 %, and 3.56 % in predicting explosion pressure (P), flame height (H), and flame velocity (V), with RMSE reductions of 38.76 %, 78.65 %, and 33.68 %, and MAE reductions of 93.25 %, 69.23 %, and 27.40 %. The analysis revealed that physical attributes like particle size play a critical role in gas/surface reactions during explosions, surpassing the impact of chemical modifications. Shapley Additive exPlanations (SHAP) analysis further demonstrated that nitrogen content (N%) exhibited a negative correlation with the explosion pressure; when nitrogen content exceeds 12.5 %, both P, H, and V are decrease markedly. Moreover, materials with larger molecular weights (>1300) can substantially reduce H and V by delaying the combustion chain reactions. SHAP reveals complex interactions among inhibitor properties and model relationships. This study not only offers a theoretical foundation for developing nitrogen/phosphorus-containing structure-controlled composite materials in inhibiting methane/coal dust explosions but also provides technical support for the design and optimization of explosion inhibitors.