Evaluating Enthalpy Production in Geothermal Reservoirs: Insights from Response Surface Methodology and Advanced Machine Learning Techniques

Abstract

Enthalpy is a key thermodynamic parameter governing energy extraction efficiency in Enhanced Geothermal Systems (EGS). Although Machine Learning (ML) has been widely applied in geothermal modeling, few studies have systematically integrated Response Surface Methodology (RSM) with ML to develop and compare multiple predictive models for enthalpy production. Using datasets from CMG STARS simulations, we developed predictive models based on RSM and four ML techniques (Random Forest, Decision Tree, XGBoost, and Support Vector Machine). A Central Composite Design (CCD) in CMG CMOST established relationships between operational parameters and enthalpy, while Particle Swarm Optimization (PSO) determined the optimal operating conditions within the investigated design space. The RSM model achieved an \(R^2\) of 0.999404, outperforming Random Forest (0.9929), Decision Tree (0.9882), XGBoost (0.9883), and Support Vector Machine (0.9888). PSO optimization yielded a maximum cumulative enthalpy of \(4.507 \times 10^{16}\) J. These results demonstrate that integrating RSM with ML provides accurate and robust predictive capability with efficient optimization of geothermal reservoir performance. This study offers a structured framework combining statistical and data-driven approaches to improve geothermal energy assessment and decision-making.