Thermal interference in aquifer thermal energy storage: Insights from stochastic surrogate modelling of well placement

Hot Temperature Aquifer Thermal Energy Storage (HT-ATES) systems have gained interest due to their high storage capacity, cost-effectiveness, and environmental sustainability. Especially in urban areas with surplus heat, geothermal district heating becomes highly beneficial.Wells in ATES must be spaced to avoid thermal interference, but urban space and regulations often constrain well placement. In this numerical study, we provide placement criteria for well doublets in urban areas. In contrast to previous model approaches, the approach optimizes the trade-off between thermal interference and well spacing by maximizing energy recovery and minimizing well distance To minimize computational costs, we designed a workflow where the full order physics-based model was replaced by a Random Forest Classifier (RFC) that predicts thermal interference with 97% accuracy accounting for technical and geological uncertainties in a typical deep HT-ATES system. The model was applied to an ATES system planned in Berlin-Adlershof. Results show that the well distance between 130 and 160 meters is optimal from the thermal and hydraulic perspective over a wide range of parameters. The global sensitivity analysis of the surrogate model reveals that influential parameters for thermal interference are well placement, aquifer thickness, flowrates, charging temperature, permeability, and porosity. By providing insights into key design priorities and enabling an explicit understanding of thermal interference, the developed RFC is an efficient tool for pre-evaluating possible ATES configurations and supporting urban planners’ decision making. • An optimal well spacing analysis which includes novel hydraulic and thermal characteristics was developed. • A modelling framework for ensemble-based estimation of thermal interferrence in HT-ATES is introduced. • The approach facilitates probabilistic assessment of thermal interference in HT-ATES doublet. • The parameter study reveals the model sensitivity and provides optimization insights.