One-step pyrolysis model integrating machine learning predictions for coal gasification simulations using multiphase particle-in-cell method

• A one-step pyrolysis model integrated with ML predictions has been developed. • The XGBoost model exhibits excellent predictive ability with an R 2 of 0.921. • The ML-integrated model is more accurate and reasonable than the empirical model. • Different one-step pyrolysis models affect gas-phase thermophysical properties. • Coal particles exhibit higher dispersion coefficients than sand particles. In Computational Fluid Dynamics (CFD) simulations of fluidized bed coal gasification, accurately describing the coal pyrolysis behavior is crucial for the accuracy of the simulation results. This study developed a one-step pyrolysis model an integrating machine learning predictions (OSPM-ML) using the XGBoost algorithm, based on 151 pyrolysis experimental datasets from coals of various ranks. The OSPM-ML demonstrated strong generalization capability, achieving a mean test R 2 of 0.921 and RMSE of 3.026. OSPM-ML was applied in MP-PIC simulations of fluidized bed gasification and compared against one-step pyrolysis models integrating experimental results and empirical model (OSPM-PE and OSPM-EM). Results showed that OSPM-ML exhibited high consistency with OSPM-PE in predicting gas–solid flow, gas-phase composition distribution, and gas-phase thermophysical properties within the reactor. Additionally, OSPM-ML accurately captured the variation of outlet products with temperature. These findings demonstrate that OSPM-ML serves as a reliable and efficient alternative, providing a robust tool for simulating fluidized bed gasification processes. Furthermore, this study investigated the thermal-physical–chemical properties of particles within the fluidized bed, offering new perspectives and insights for advancing the understanding of fluidized bed coal gasification processes.