Modeling and optimization of borehole thermal energy storage systems using physics-based neural networks

Borehole thermal energy storage (BTES) systems are critical components in the decarbonization of district heating networks as they enhance operational flexibility by seasonally storing thermal energy. While various modeling approaches for BTES exist, they are typically unsuitable for optimization problems and model-based control design. In this work, we propose a novel modeling approach that leads to a physics-based neural network surrogate model of the BTES temperature dynamics, capturing the fundamental dynamics with sufficient accuracy, while maintaining a relatively low complexity that makes it suitable for deployment in operational optimization or control algorithms. Specifically, we examine a standard BTES system that combines multiple heat sources, heat pumps, and storage buffers, detailing the associated mass flows and temperatures. We utilize a Python-based operational optimization process for the theoretical system setup using Pyomo and demonstrate that our modeling approach enables accurate optimizations over planning horizons of up to one year with a sample time of one hour. The new modeling approach significantly improves prediction accuracy across the relevant system states, with mean absolute errors reduced by approximately one-third compared to a single-capacitance model identified with the sparse identification framework SINDy.