Experimental and machine learning associated study into the spontaneous combustion susceptibility of the steam coals

OpenAlex · English

This study proposes an integrated experimental – machine learning (ML) framework to assess the spontaneous combustion susceptibility of imported steam coals (ISC). The approach combines standardized laboratory measurements with multiple advanced ML algorithms to predict key liability indices, including crossing point temperature (CPT), average temperature increase (ATI), and the Feng – Chakravorty – Cochrane (FCC) index. A dataset of thirty-four coal samples was generated through comprehensive physicochemical characterization and corresponding susceptibility testing. Model performance was systematically evaluated using statistical metrics, and feature importance analysis was conducted to determine the influence of coal properties on combustion behavior. Results indicate that the support vector machine (SVM) consistently delivers the best performance, achieving an R2 of 0.87 for almost all indices. For CPT, SVM and gradient boosting decision tree (GBDT) achieved R2 values of 0.87 and 0.77, respectively. In ATI prediction, partial least squares (PLS) and SVM yielded R2 values of 0.87 and 0.83, while for FCC, SVM and Gaussian process regression (GPR) reached R2 values of 0.87 and 0.85. Overall, the proposed framework provides a reliable and practical tool for evaluating spontaneous combustion risk and improving coal handling safety.