Co-optimization of well schedule and conformance control parameters assisted with Transformer-LSTM for CO2-EOR and storage in oil reservoirs

The injection of CO2 into oil reservoirs offers dual benefits by enhancing oil recovery and facilitating CO2 storage to address global warming. However, gas channeling frequently occurs during the CO2 flooding process due to reservoir heterogeneity or improper well control parameters, which reduces both CO2 sweep coefficient and storage capacity. Although optimizing well scheduling or conformance control parameters can mitigate gas channeling, these approaches are often time-consuming due to computationally intensive numerical simulations. This paper proposes a framework that integrates deep learning algorithms with numerical simulations to rapidly maximize net present value (NPV) and CO2 storage capacity through the collaborative optimization of multi-stage well scheduling and conformance control parameters. A Transformer-LSTM model is iteratively trained and updated throughout the optimization process to accelerate algorithm convergence and enhance the efficiency of the co-optimization framework. The validity and superiority of the co-optimization framework were tested on both a 3D synthetic model and a real reservoir model. The results show that, compared to traditional strategies that optimize only well scheduling or conformance control parameters, the co-optimization strategy increases NPV by 12.15–13.52 %, improves CO2 storage by 6.05–21.45 %, and reduces optimization time by about 70 %. This study provides an accurate and efficient framework for optimizing CO2-EOR and storage parameters, offering valuable guidance for integrated development and supporting reservoir engineering decision-making.