Study on multi-field coupling numerical simulation of in-situ pyrolysis of tar-rich coal assisted by machine learning

In the context of the "double carbon" goals and global energy structure adjustments, underground in-situ pyrolysis technology for tar-rich coal has become an important direction for alleviating China's dependence on foreign tar and gas due to its high efficiency and low carbon footprint. To address the challenges posed by complex underground in-situ conditions, difficult pyrolysis process intensification and optimization of process parameters, this study takes tar-rich coal from Yulin in northern Shaanxi as the research object. A multi-field coupled numerical model integrating thermal, flow and chemical reaction processes was constructed based on Darcy's seepage law and pyrolysis reaction kinetics equations, and the dynamic evolution of temperature field, pressure field, fluid seepage field and product generation was systematically simulated. The influence of heat carrier temperature (550–800 °C), injection pressure (4–8 MPa), coal seam permeability (50–300 mD, 4.94×10^-14–2.96×10^-13 m²) and wellhead deployment schemes (single well, double well and four well heat injection) on pyrolysis process was systematically studied by combining multi-field coupled numerical simulation with random forest machine learning, and an intelligent optimization model was constructed to minimize operating cost and reach the standard of conversion rate. The results indicate that an optimal efficiency range exists at pyrolysis temperatures of 650–700 °C, with a conversion rate of 92.88% achieved after heating at 700 °C for 100 days. The energy consumption increases gradually. The injection pressure of 8 MPa can expand the high-concentration area of light tar and gas products by 30% compared to that of 4 MPa, and the uniformity of the temperature field is significantly improved. Under the condition of 300 mD (2.96×10^-13 m²) permeability, the heat transfer efficiency of the pyrolysis reaction zone extending to the deep part of the coal seam is obviously enhanced. In the multi-well heat injection scheme, the deployment of four wells can significantly improve the uniformity of the pyrolysis temperature field for 150 days, and the distribution range of products is more obvious than that of a single well. Based on 820 sets of simulation data, the predictive determination coefficients of the random forest model for temperature and conversion rate are 0.988 8 and 0.997 3, respectively. The combination of injection temperature of 589.47 °C, injection pressure of 4.0 MPa, permeability of 76.3 mD (7.53×10^-14 m²) and heating time of 189.5 days can be optimized, and the cooperative optimization of 90.47% conversion rate and operating cost can be realized. The research results provide a theoretical basis for process parameter design of tar-rich coal in-situ pyrolysis, and establish a data-driven optimization framework for engineering decision-making.