A fast and accurate Data-Driven Model for estimating the production temperature of High-Temperature Aquifer Thermal Energy Storage

Abstract

High-Temperature Aquifer Thermal Energy Storage (HT-ATES) has the potential to significantly increase the renewable heat share in heating systems. However, HT-ATES has not been implemented in the current energy system models because the widely applied numerical models for HT-ATES are computationally expensive. This leads to a lack of HT-ATES assessment from an energy system perspective. Therefore, an accurate and computationally efficient model that is widely applicable is needed to facilitate such implementation. This research aimed to develop a novel data-driven model that generates the temperature profile of an HT-ATES accurately and computationally efficiently. A trained machine learning algorithm predicts the recovery efficiency for an HT-ATES system, which, combined with other parameters, enables a nearest neighbor search to identify a suitable temperature profile. As a result, the temperature profile generated by the data-driven model has a root mean square error of 1.22 °C compared to the numerical model output. This error was shown to be larger for lower recovery efficiency values compared to higher values. The machine learning algorithm used to predict the recovery efficiency has a root mean square error of 1.45 percentage points. The data-driven model has a computation time of less than half a second, which is more than 180,000 times faster than the numerical model that was used to generate the data. This model is, therefore, suitable for integration in larger energy system models.