Machine learning-based prediction and experimental validation of the pyrolysis product distribution in tar-rich coals

BackgroundTar-rich coals serve as an important coal-based oil and gas resource in China, while their pyrolysis product distribution is governed by the coupling effects of coal properties and reaction conditions. Therefore, rapidly identifying the pyrolysis product distribution patterns holds great significance for the resource evaluation and experimental design of tar-rich coals.Methods Existing studies on tar-rich coals suffer from the insufficient integration of exclusive data and limited synergistic prediction capacities for multiple products. To address these issues, this study constructed a dedicated dataset involving proximate analysis, ultimate analysis, elemental molar ratios, maceral composition, and pyrolysis conditions by systematically collecting experimental data from associated literature. Subsequently, a feedforward neural network model for multi-target regression was established for the basic prediction of pyrolysis product yields. Accordingly, this study identified primary factors controlling the pyrolysis product distribution of tar-rich coals by combining Van Krevelen diagrams, Spearman correlation analysis, and feature contribution analysis based on SHapley Additive exPlanations (SHAP) values. Finally, through pyrolysis experiments under 400‒600 ℃, this study validated the model’s capability to reproduce the major temperature response patterns of the pyrolysis products of the tar-rich coals.Results and ConclusionsThe established feedforward neural network model effectively reproduced the typical temperature response patterns of the pyrolysis products of tar-rich coals. Specifically, with an increase in the pyrolysis temperature, the char yield decreased, the gas yield increased, the water yield increased overall, and the tar yield increased initially and then decreased. This model exhibited an average coefficient of determination (R2) of 0.89 and an average root mean square error (RMSE) of 1.53 on the testing set. Among primary influencing factors, coal properties, particularly carbon content, vitrinite content, and volatile constituent content, generally produced more significant impacts on the product distribution than external operating parameters like heating rate and particle size. Experimental validation demonstrates that good agreement existed between predicted and experimental gas and char yields, while the tar yield was slightly overestimated under high-temperature conditions. Therefore, future improvement should focus on sample coverage and secondary pyrolysis under high-temperature conditions. The results of this study reveal the basic distribution patterns of the pyrolysis products of tar-rich coals and verify the capability of the established feedforward neural network model to reproduce the major temperature response trends of the pyrolysis products of tar-rich coals. These results will provide references for the resource evaluation, pyrolysis experimental design, and product control research of tar-rich coals.