Machine Learning Assisted Semi Real Time Geomechanical Risk Evaluation During Methane Hydrate Production

Abstract Geomechanical risks such as sanding are common during methane hydrate production, posing significant engineering challenges. These risks are often associated with changes in pressure, porosity, and saturation along the wellbore radial direction, which in turn affect the strain state at the wellbore. In this study, a machine learning-based surrogate model is established to predict wellbore strain responses under various production scenarios, including different drawdown pressure, depletion fronts, dissociation fronts, and sanding events. The model is trained using synthetic data that simulate realistic production conditions and geomechanical responses. By integrating strain data with the surrogate model, actual production conditions can be inversely determined in semi real time, allowing for the evaluation of geomechanical risks such as the advancement of depletion and dissociation fronts, as well as the onset of sanding. This methodology enables flexible and timely geomechanical risk assessment during methane hydrate production, especially for the sanding front prediction and sanding volume calculation.