A time-series forecasting model-based optimization approach for well-doublet system in geothermal reservoirs under geological uncertainty

Optimization of well placement and control parameters is essential for efficient and sustainable geothermal energy exploitation. However, the computational burden induced by simulation with multi-physical processes and geological uncertainty presents a significant challenge. Additionally, uncertainty can substantially impact the optimization results. This study proposed an optimization approach based on a hybrid time-series forecasting surrogate model, which is trained on a large numerically-generated dataset to effectively predict reservoir responses. The objective function of the optimization is the net profit value (NPV) over an exploitation cycle, subject to the constraint of sustainability standard. The genetic algorithm-based optimization strategies including stochastic optimization and robust optimization, were employed to provide the optimal solution with high-expectation or high-lower-bound NPV, in risk-neutral or risk-averse scenarios, and geological uncertainty was propagated through Monte-Carlo simulation. A comparative analysis of methodological approaches was also conducted. A case study on the Tsinghua doublet system demonstrates that the proposed approach significantly enhances decision-making of well control, by accounting for both geological uncertainties and scenario-based optimization requirements, providing a valuable tool for geothermal reservoir management.