Probabilistic assessment of the thermal performance of low-enthalpy geothermal system under impact of spatially correlated heterogeneity by using XGBoost algorithms

Low-enthalpy geothermal energy represents a widely accessible renewable resource. However, efficient heat extraction in such systems remains complex due to uncertainties associated with spatially correlated reservoir heterogeneity. This study presents a computational framework that integrates numerical simulations with data-driven modeling to analyze the impact of reservoir heterogeneity on thermal performance. Initially, 6000 simulations were conducted on heterogeneous models, yielding 5866 valid results to train and validate a surrogate XGBoost model. SHAP analysis was utilized to systematically assess the influence of reservoir heterogeneity on thermal performance. To quantify the likelihood of not meeting design specifications, a failure probability was introduced and computed based on 64,000 additional predictions from the XGBoost model. Results suggest a generally positive correlation between porosity and all thermal performance indicators. High levels of reservoir heterogeneity are likely to decrease thermal breakthrough time, thermal production lifetime, and production capacity. Feature importance analysis identified mean porosity as the most significant variable, followed by porosity at injection and production well. In highly heterogeneous reservoirs, uncertainties can cause intricate variations in performance metrics. In cases with limited geological data, the failure probability metric offers a practical means for rapidly evaluating thermal performance during early-stage design.