Integration of Domain Knowledge and Data-Driven Modeling Evaluation Process for Predicting Minimum Miscible Pressure of CO₂-Oil Systems in CCUS

Carbon dioxide-enhanced oil recovery (CO2-EOR) is the main application of carbon capture, utilization, and storage (CCUS) in oil and gas field development, and the modeling and evaluation process of the minimum miscible pressure (MMP) of the CO2-crude oil system is very important for CO2-EOR projects. Unlike previous studies that often used low-dimensional incomplete data, this research utilizes high-dimensional nonlinear full-component tabular data of injected gas and crude oil. The mixed screening method is employed to identify significant features. This paper addresses the issues in traditional intelligent MMP prediction models, which tend to overlook the influence of all fluid components, lack rigorous feature screening, and rely on a single evaluation method. The WOA-DBN intelligent model is built using domain knowledge. The results of various models are compared and assessed using three methods: model accuracy evaluation, learning curve evaluation, and single control variable evaluation. The findings demonstrate that WOA-DBN, with its high-dimensional nonlinear tabular data processing capability, outperforms other methods for predicting MMP values. The TIC value is 0.0266, the MRE value is 0.0553, and the R-2 value is 0.9792. The learning curve indicates that WOA-DBN effectively learns from tabular-based data. Additionally, we think that the typical intermediate heavy hydrocarbon components (C-5 and C-6) in crude oil should be taken into account independently, alongside components C-2, C-3, C-4, C-6, and C-5. Ignoring the influence of any component can lead to biased analyses. The most crucial variables also include the reservoir temperature (T-R) and the mole percentages of N-2 and C-1 in the injected gas. The prediction model is highly applicable and follows specific mechanism principles. Any high-dimensional nonlinear tabular data in the field of fossil energy can be handled by the modeling and evaluation technique in this work, which also serves to direct field production and a related theoretical study.