Integration of at-scale field observations and application of surrogate modelling strategies for optimal design of ATES systems in the Sherwood Sandstone aquifer

Aquifer Thermal Energy Storage (ATES) systems use reversible abstraction and injection in combination with warm and cold wells to provide efficient heating and cooling solutions at scales up to ~0.5 MW per installation. It has been shown that these systems are able to improve the efficiency of thermal installations, but rely on an appropriate design, especially when several ATES systems share the same aquifer. In this work, we combine groundwater flow and heat transport numerical models with optimization frameworks to investigate the optimal distribution of wells for avoiding system interferences and improving recovery efficiency. To that end, we employ hypothetical modelling scenarios based on geological properties of the Sherwood Sandstone bedrock aquifer as one of the main potential targets for the development of ATES systems in the UK. Some of the available information is acquired from activities at the UK Geoenergy Observatory (UKGEOS) in Cheshire, which is under construction and will be equipped with a range of technologies and monitoring sensors for research, training, and on-site experiments. The Observatory will be open to industry and the research community to evaluate technological options for shallow geothermal use and energy storage and to gain a detailed hydraulic and thermal characterization of the Sherwood Sandstone. As practical application, we present an approach for the optimal design of the installation of multiple doublets that consider various spatial features as decision variables. The benchmark solution is provided by a simulation-optimization framework that uses a direct coupling of the groundwater flow and heat transport numerical model with an evolutionary algorithm. This approach is typically hampered by increased computational cost due to the time-intensive numerical simulations and the thousands of objective function evaluations required until convergence of the evolutionary algorithm is achieved. Therefore, we also investigate a lower computational resource strategy by applying surrogate-assisted optimization methods which are either embedded in the operations of the evolutionary algorithm or utilize an adaptive-recursive framework. The performance of the surrogate-based optimization method is assessed via several independent optimization trials and for different computational budgets. The ability of the surrogate-based optimization frameworks to approximate a near global solution is compared against the benchmark solution.