Data-driven machine learning models for predicting deliverability of underground natural gas storage in aquifer and depleted reservoirs

Subsurface parameters, like geological formations and fluid dynamics, are vital for evaluating connectivity in aquifers and depleted reservoirs. Accurate models predicting Underground Natural Gas Storage (UNGS) deliverability are crucial for stakeholders due to demand-supply inconsistencies but remain complex and underexplored. Therefore, in this study, machine learning (ML) models were employed, particularly Artificial Neural Networks (ANNs) optimized with; the Levenberg-Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG), to predict the deliverability of UNGS in the aquifer and depleted reservoirs. The prediction capacities were assessed using 2084 field storage data points from 78 operational reservoirs in the United States. Performance evaluation based on root mean square error (RMSE) and coefficient of determination (R) showcases the superior predictive precision of the BR algorithm, accomplishing RMSE values between (0.010–0.0496) and R values of (0.909–0.997). Moreover, the Regression Error Characteristic (REC) curve and Taylor diagram both indicate that the BR model significantly outperforms the LM and SCG models in terms of accuracy and reliability. Furthermore, validation has been performed using datasets from active aquifers and depleted reservoirs in 2024, which showed good accuracy and reliability. Hence, the proposed ML models effectively forecast UNGS deliverability and provide clear insights for informed decision-making.