Generalized regression neural network-assisted synergistic optimization of oil recovery, heat extraction, and carbon sequestration in deep reservoirs.

This article focuses on the development and validation of a novel GRNN-assisted Surrogate Optimization Algorithm (GSOA) designed to optimize CO2-enhanced oil recovery (CO2-EOR), geological sequestration, and geothermal energy co-production in deep reservoirs. The GSOA integrates a Generalized Regression Neural Network (GRNN) classifier for hierarchical prescreening with a Radial Basis Function (RBF) surrogate model and Differential Evolution (DE) to efficiently address high-dimensional optimization challenges, improving search space diversity and convergence speed. Applied to a numerical model of a moderate-to-high temperature reservoir in Xinjiang, China, GSOA outperformed conventional surrogate-assisted optimization by increasing cumulative oil production by 3.9%, CO2 storage by 5.2%, heat extraction by 0.9%, and net present value (NPV) by 3.2%, while reducing computational time by nearly half. The study demonstrates GSOA’s potential as an effective tool for synergistic resource development and carbon-neutral strategies, though its broader application depends on site-specific reservoir and CO2 availability conditions.