限られたサンプル条件下でのCO2貯留と原油増進回収性能の評価

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a novel RBF-based surrogate method for CO2-EOR-storage evaluation that outperforms DNN under limited data, useful for similar reservoir modeling studies.

🏢実務担当者:Offers a practical tool for screening CO2 injection strategies and assessing economic feasibility in mature oilfields, aiding CCUS project planning.

🏛政策担当者:Highlights the economic conditions under which CO2-EOR becomes viable, informing policy support for CCUS deployment.

Under China’s carbon peaking and carbon neutrality goals, CO 2 capture, utilization, and storage provides an important pathway for simultaneously improving oil recovery and reducing carbon emissions in mature oilfields. For heterogeneous conglomerate, tight, and mature reservoirs , rapid evaluation of CO 2 injection schemes is challenging because high-fidelity numerical simulation is computationally expensive and available training samples are often limited. This study develops a radial basis function (RBF) network-based surrogate, which is a purely data-driven table lookup interpretation model, to evaluate the coupled performance of CO 2 enhanced oil recovery and geological storage. A total of 162 high fidelity simulation cases were generated through full-factorial design, covering different water-cut stages, permeability levels, heterogeneity conditions, and injection strategies, including continuous CO 2 injection and five water-alternating-gas ratios. Monthly cumulative oil production and cumulative CO 2 storage were used as prediction targets. A thin-plate spline RBF interpolator was constructed and compared with a deep neural network benchmark trained using the same dataset and testing strategy. Blind testing shows that the RBF model achieves endpoint R 2 values of 0.8952 for cumulative oil production and 0.9402 for CO 2 storage, while the corresponding DNN values are only 0.5288 and 0.6329. For full monthly time-series prediction, the RBF model obtains R 2 values above 0.96 for both targets, clearly outperforming the DNN in accuracy, stability, and smoothness of cumulative prediction curves. The RBF surrogate was further coupled with a net present value module to construct decision matrices across oil-price and CO 2 -cost conditions. Results indicate that continuous gas injection is favorable in relatively homogeneous reservoirs under low CO 2 -cost conditions, whereas WAG schemes become increasingly advantageous as heterogeneity or CO 2 cost increases. In strongly heterogeneous reservoirs, WAG schemes dominate most of the economic decision space because they improve sweep efficiency and reduce early gas breakthrough risk. The proposed workflow provides a data-efficient and computationally practical tool for early-stage CO 2 -EOR-storage screening and injection-strategy selection.

• semanticscholar https://doi.org/10.1142/s2972379526500055first seen 2026-09-05 05:34:23