2021 Volume 77 Issue 1 Pages 21-34
For commercial scale deployment of Carbon dioxide Capture and Storage (CCS), multiple wells will be required to inject a large volume of CO2 into a deep reservoir. Multiple well placement is the key to realize required injectivity and capacity of the reservoir. Recent studies showed that an optimization tool combining metaheuristics including the generic algorithm with numerical simulators is effective for optimizing many well parameters such as injection and production rate and well locations simultaneously. However, calculation time for the optimization could be a problem because the tool requires some thousands of simulations. Especially, applying the tool for a model in a real CCS project that reflects heterogeneity of physical properties is not realistic. In this research, we proposed a new method for mitigating the problem by leveraging parallel computation technique and supercomputer. The method was applied for an optimization tool combining a parallel reservoir simulator TOUGH2-MP with a meta-heuristics Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The practicality of the tool on a supercomputer Oakforest-PACS (0.55 million CPU cores) was investigated through case studies of well placement optimization in a heterogeneous reservoir model. As a result, the tool could find optimum solution within the realistic time (several weeks). The result suggests that the tool can contribute to optimize well placement in a real CCS project.