This paper proposes a multiscale numerical simulation and parallel computing algorithm for the ASP (alkali-surfactant-polymer) enhanced oil recovery technology. The algorithm establishes a physicochemical mechanism model incorporating alkali consumption kinetics, polymer shear degradation, and chromatographic separation effects, achieving macroscopic-microscopic dynamic mapping through the volume averaging method. Concurrently, a GPU-CPU-FPGA co-computing platform was designed. By integrating adaptive mixed-mesh technology with machine learning acceleration, the simulation time per run was reduced to under 2 hours. Furthermore, a task scheduling strategy with dynamic load balancing and a fault-tolerant mechanism were developed to support minute-level adjustments of injection parameters. Validation tests using a strongly heterogeneous conceptual model demonstrated high consistency between the algorithm's simulation results and commercial software, with cumulative oil production prediction errors below 2.5%. Through combined acceleration using GPU/FPGA heterogeneous parallel processing and machine learning proxy models, simulation time was drastically reduced from 3 days to under 2 hours, achieving a 40-fold efficiency improvement. Field application and parameter optimization tests demonstrate that this algorithm effectively fits actual production data and rapidly completes predictive simulations for different injection schemes, providing robust decision support for on-site optimization adjustments in oilfields.
Qian Li (Sun,) studied this question.