A Physics-Informed Transformer Framework for Spatio-Temporal Modeling and Energy-Carbon Tradeoffs in CCUS-EOR

Abstract Carbon Capture, Utilization, and Storage for Enhanced Oil Recovery (CCUS-EOR) plays a vital role in achieving carbon neutrality while ensuring energy and economic development. Although process modeling and scheme optimization are essential for CCUS-EOR, existing methods often have difficulty in simultaneously describing the complex spatiotemporal behavior of reservoirs and performing effective optimization design. Therefore, they cannot fully meet the task requirements of energy-carbon integrated coupled prediction and optimization. To address the challenge, we propose an intelligent decision-making framework that integrates a Spatio-Temporal Physics-Informed Transformer (STPIformer) with an Adaptive Operator-Controlled NSGA-II (AOC-NSGA-II) algorithm. By embedding CO2–oil–water three-phase flow partial differential equations (PDEs) into a Swin Transformer backbone and incorporating WAG-aware temporal encoding module and spatio-temporal attention mechanisms, STPIformer enables high-fidelity prediction of reservoir dynamics, including oil migration and CO2 plume evolution. The AOC-NSGA-II algorithm further enhances multi-objective optimization by adaptively balancing energy-carbon trade-offs among cumulative oil production, CO2 geological storage, net present value (NPV), and carbon emissions. Field-scale 3D model validation on the Tuha oilfield demonstrates that the proposed method significantly enhances spatio-temporal prediction accuracy and generalization performance, while delivering optimization schemes that satisfy underlying physical constraints and balance environmental and economic objectives for task-specific deployment. This study provides a robust modeling and optimization framework to support the large-scale deployment of CCUS-EOR.