Modeling of methane adsorption capacity in shale gas formations using white-box supervised machine learning techniques

Energy demand is increasing worldwide and shale gas formations have gained increasing attention and have become crucial energy sources. Therefore, accurate determination of shale gas-in-place (GIP) is vital for a successful production plan. Since most of the gas in shale formation is in the form of adsorbed gas, the determination of methane adsorption capacity is a very important task. In this study, two rigorous data-driven techniques, namely gene expression programming (GEP) and group method of data handling (GMDH), were utilized to provide accurate and reliable explicit mathematical expressions for predicting methane adsorption. For this purpose, a comprehensive database involving 352 data points was gathered from the literature. Pressure, temperature, moisture, and total organic carbon (TOC) were employed as input variables for the implemented correlations. Results indicate that both correlations can provide accurate predictions. However, the GEP-based correlation exhibits more reliable predictions for methane adsorption with a correlation coefficient of 0.9837. Moreover, it was shown that GEP-based correlation can accurately predict the variation of shale gas capacity for the alteration of inputs. Further, it was revealed that methane adsorption is highly dependent on moisture value, while temperature, TOC, and pressure are the most influential variables after moisture. The results of this study shed light on the power of GMDH and GEP modeling approaches, and show that these models can be employed to provide accurate and simple-to-use correlations for estimating methane adsorption in shale gas formations.