A Robust General Physics-Informed Machine Learning Framework for Energy Recovery Optimization in Geothermal Reservoirs

Energy extraction from the Enhanced Geothermal System (EGS) is highly dependent on the transmissivity of fractures. However, due to the heterogeneity and complex multi-physics nature, high-fidelity physics-based forward simulation can be computationally intensive, creating a barrier to efficient reservoir management. A robust and fast optimization framework for maximizing the thermal recovery from EGS is needed. We developed a general reservoir management framework which is combining a low-fidelity forward surrogate model (f_l) with gradient-based optimizers to speed up reservoir management process. thermohydro-mechanical (THM) EGS simulation model is developed based on the finite element method. We parameterized the fracture aperture and well controls and ran the THM model to generate 2500 datasets. Further, we used two different deep neural networks (DNNs) with the datasets to predict the dynamics of pressure and temperature, and this ultimately becomes the f_l for calculating the energy production. Instead of performing optimization workflow with large amount of simulations from f_h, we directly optimize the well control parameters based on geological input to the f_l. As the f_l can reach the high accuracy with fast prediction, also it is differentiable, gradient-based optimization was utilized to find the maximized total energy production optimum with temperature constraint at producer. Based on the simulation datasets, we evaluated the fracture aperture and temperature evolution and demonstrated that the spatial fracture aperture distribution dominates the thermal front movements. The fracture aperture expansion is highly correlated with temperature change inside of the fracture, mainly from thermal stress changes. Compared to the full-fledged physics simulator, our forward surrogate model based on DNN not only provides a computational speedup of around 1500 times, but also brings a high R² value about 99% in predicting subsurface responses. With the aid of the efficient forward model f_l, gradient-based optimizers show efficient optimization with 10-68 times faster than the derivative-free global optimization method. The proposed reservoir management framework shows both efficiency and scalability, which enables each optimization process to be executed within half a minute.