Defining Injection Wells Location in CCS Projects Using Bayesian Optimization

Summary Deploying CCS on a large scale demands a meticulously planned and executed storage Field Development Plan (FDP) which is based on complex optimization techniques, utilizing numerous CPU time demanding computer simulations. Issues such as optimal well locations and appropriate injection/production strategies which maximize storage while preventing CO2 migration beyond designated boundaries and incorporate economic factors and market fluctuations, need to be addressed. Bayesian optimization, originating from the Machine Learning area, arises as the proper selection to allow the optimization of expensive cost functions as is the case with CCS simulation. It provides the probability density function (pdf) of the objective function for all possible solutions, out of which the one exhibiting maximum probability needs eventually to be realized. We demonstrate the Bayesian approach to optimize the location of CO2 injection and brine withdrawal wells in CCS applications. Firstly, various objective function forms, accounting for different CCS tasks, along with their accompanying constraints are considered. Subsequently, the degree to which this method consistently leads to considerably satisfying results at a limited number of storage simulations is shown. Finally, we investigate the statistical interpretation of the obtained probability distribution solution as well as the incorporation of existing knowledge.