Machine learning for product distribution prediction of one-step pyrolysis model of coal

Pyrolysis is the first step in the coal fluidized bed thermo-chemical conversion process, and its product distribution directly affects subsequent gasification and combustion processes. Accurate prediction of the one-step pyrolysis product distribution under different operating conditions and coal ranks is of significant importance for numerical simulations. In this study, a machine learning (ML) approach was employed to predict the pyrolysis product distribution in the one-step pyrolysis mode with 151 experimental datasets from various coal ranks. It was found that the XGBoost model exhibited the best overall predictive performance. After pruning and regularization optimization, the model's predictive capability was further enhanced, achieving an R² of 0.921 and reducing the RMSE to 3.026. Compared with empirical model, the ML model produced predictions that were more consistent with experimental data across three different coal ranks and successfully captured the temperature-dependent variations in pyrolysis product distributions. Regarding input feature importance, carbon (C), hydrogen (H), and oxygen (O) contents, along with temperature (T), were identified as the most critical factors influencing the distribution of three-phase pyrolysis products and gas composition. Additionally, particle diameter (d_p) was found to play a significant role in predicting the concentrations of CO₂, H₂, and CH₄. Furthermore, this study provides insights into the application of the one-step pyrolysis model and the adjustment of product distribution strategies, as well as theoretical guidance for optimizing fluidized bed reactor operation.