Design of low carbon CO2-alternative-water injection processes using a machine-learning-assisted multi-objective optimization protocol

Large-scale CCUS projects can generate considerable CO2 emissions during operational processes. However, very little attention has been paid to the CO2 emissions associated with the injection process itself for field scale CO2 enhanced oil recovery (EOR) projects. A notable contribution of this study is the incorporation of CO2 intensity into multi-objective optimization framework employing a thermodynamic variable exergy. This CO2 emission related metric is considered alongside cumulative oil production, CO2 storage capacity and net present value (NPV) to obtain a comprehensive strategy. A hybrid machine learning model that integrates deep neural networks (DNN) with support vector machine (SVM) is trained as the surrogate to the high-fidelity numerical simulator. Coupling with the NSGA-II optimizer, Pareto optimum set can be structured with non-dominating solutions considering the aforementioned objectives. The proposed optimization workflow is implemented to a benchmark CO2 water alternative gas (WAG) project and investigate the compromise relationships among objectives. A weighted scoring criteria is employed to screen the Pareto optimum set and identify the most practical scenarios. Results show that compared with the original development strategy, the best scoring solution achieves a 37 % increase in cumulative oil production and enhances CO2 storage by 38.4 million metric tonnes. Moreover, average CO2 emissions per kilogram of CO2 injected or per cubic meter of water injected are reduced from 0.053 kg to 0.029 kg achieving a 45 % reduction, with the NPV increased by 47.36 %. The proposed optimization strategy offers a practical solution toward low-carbon and efficient field development.