Mamba-attention surrogate-assisted optimization framework for well placement and dynamic schedule in CO2 enhanced oil recovery and storage

CO2 enhanced oil recovery (EOR) significantly boosts both oil recovery and carbon sequestration. Well placement and operational strategies are crucial factors that influence the effectiveness of CO2 flooding. Traditional optimization methods often rely on compositional numerical simulations, which are costly and inefficient. This paper proposes a new framework for the fast, simultaneous optimization of well placement and scheduling. First, a Uniform Gas Displacement Index (UI) is introduced, which utilizes relative gas saturation deviations from regional averages across the reservoir grid. This index quantitatively characterizes the uniformity of fluid distribution, accurately capturing gas front movement and displacement heterogeneity, thus providing reliable support for optimization decisions. Second, a multi-objective optimization framework that integrates the Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is developed to maximize both UI and Net Present Value (NPV), aiming to balance CO2 sweep efficiency with economic returns. Third, the improved Mamba-Multi-Head Attention (Mamba-MHA) surrogate model is developed to replace computationally intensive numerical simulations, thereby significantly reducing computational costs. Validation through a 3D synthetic case and application in the complex DHS reservoir demonstrates that the framework outperforms traditional approaches. Notably, the framework shows approximately 80% improvement in optimization efficiency. Under identical computational constraints, it achieves superior economic returns while effectively controlling gas channeling. This research advances uniform displacement theory and offers a robust technical solution for balancing economic performance with environmental protection in reservoir development and CO2 storage.