During CO2-EOR implementation, CH4 in the reproduced CO2-rich mixture are included in the recycle injection gas. However, improper injection gas composition and operation parameters can reduce the flooding performance.In this study, a machine-learning assisted framework combining Convolutional Neural Networks-Gate Recurrent Unit (CNN-GRU) and Non-dominated Sorting Genetic Algorithm II (NSGA-II) was proposed to obtain the optimal parameters for simultaneous maximum oil production, carbon storage efficiency, and net present value. A case study from the CO2-EOR project in the H59 block of Jilin oilfield, China, was carried out. Base cases were established to evaluate the effects of reinjection CH4 on flooding performance.Results show that the built CNN-GRU model has high prediction accuracy and computational efficiency, and thus can be used as an alternative tool to the reservoir simulator. The proposed framework can find the optimum parameters to improve oil recovery, carbon storage and net present value. The co-injection of CH4 and CO2 improves oil production, and carbon storage performance, but reduces the net present value. This work provides engineers with multiple strategies for decision-making to simultaneously promote flooding performance with CH4 being a co-injectant in CO2-EOR projects.
Read full abstract