Modeling geothermal reservoirs permeability based upon NMR laboratory data

Accurate permeability characterization is crucial for the efficient and sustainable development of geothermal resources. However, conventional methods like well testing and core analysis are often expensive and fail to capture the complex, heterogeneous nature of geothermal reservoirs. While Nuclear Magnetic Resonance (NMR) logging provides valuable insights into pore structure, its traditional permeability models are often unreliable in high-temperature, high-salinity geothermal environments. A novel data-driven methodology is introduced for modeling permeability in geothermal reservoirs by integrating Nuclear Magnetic Resonance (NMR) laboratory measurements with advanced machine learning algorithms. The approach employs a curated dataset of geothermal core samples, utilizing porosity, logarithmic mean transverse relaxation time (T2lm), and mode transverse relaxation time (T2mode) as predictive features across multiple learning models. Outlier detection was conducted using the Leverage technique, while model reliability was validated through K-fold cross-validation. Among the tested algorithms, the Decision Tree model demonstrated superior performance, yielding the highest coefficient of determination (R 2 ) and the lowest error metrics. Sensitivity analysis further revealed porosity as the most dominant factor influencing geothermal permeability. The findings validate the utility of using ensemble soft computing to boost the accuracy of permeability prediction, presenting a valuable and affordable alternative to traditional techniques. Our findings bridge the gap between core analysis and computational modeling, paving the way for more accurate geothermal reservoir characterization and optimization. • Accurate prediction of permeability in geothermal reservoirs • Sensitivity analysis of effective parameters using SHAP and relevancy index • Outlier detection using a renowned approach