Minimizing number of injection wells by binary optimization for geological carbon dioxide sequestration

Abstract

Geological Carbon dioxide Sequestration (GCS) is a viable technique to reduce CO2 volume emitted to the atmosphere by injecting it into deep geological reservoirs through wells. Commercial-scale GCS needs multiple injection wells to store a large amount of CO2 exceeding some 1Mt/year. The cost for building these injection wells directly affects the project cost, therefore, the number of wells should be optimized. At the same time, feasibility of the required CO2 injection volume into the reservoir has to be evaluated. Recently, a tool has been developed by combining a stochastic optimization algorithm such as GA and PSO with a reservoir simulator for automatically optimizing the well placement. The developed tools are formulated as a continuous or mixed-integer optimization problem with many parameters such as injection/production rates and well locations as design variables. The number of wells is implicitly optimized by the injection/production rates. For example, injection wells are built when their rates are more than a user-defined one. However, proper initialization for the approach is difficult. In addition, some cases might need only the number of wells rather than precise injection rates. In this study, we propose a different approach in which optimization problem is formulated as a binary optimization whose design variables can take 0 (no well) or 1 (building an injection well) in the designated locations. Developed is a new optimization tool combining a Population-Based Incremental Learning (PBIL) for an optimization algorithm with a CO2 flow simulator TOUGH2-MP. Performance of the tool is demonstrated through case studies in a hypothetical reservoir model with an anticline structure.