Operational strategy optimization of an existing ground source heat pump (GSHP) system using an XGBoost surrogate model

Ground source heat pumps (GSHPs) are experiencing a surge in popularity due to their efficient use of geothermal energy for building heating and cooling. Optimizing GSHP systems is crucial for maximizing heat extraction efficiency. However, existing research primarily focuses on design-stage optimization, neglecting the potential of improving operational strategies in already deployed systems. This study proposes a novel method for optimizing the operational strategy of existing GSHPs with machine-learning-based surrogate models. Our approach considers both demanded heat extraction and the variability of GSHP operation, including running and idle times. First, a numerical model is established with a closed-loop procedure replicating real-world GSHP operation, capturing dynamic inlet temperature changes. The Latin hypercube sampling is subsequently employed to generate sufficient data for XGBoost model development, where a designed disassembly method is proposed to shed light on the problem of a fixed number of input parameters. The operational strategy of the GSHP system is then implemented to offer solutions for both regular and irregular modes. The effectiveness of the method is validated using 40 days of real-case monitoring data. The results demonstrate that the surrogate model achieves high accuracy, with an absolute temperature difference error below 0.1 K and a heat extraction absolute percentage error (APE) less than 2.0 %. Additionally, the optimized operational strategy reduces total operation time by 2-9 % solely through scheduling adjustments, showcasing its significant practical potential. This study presents an efficient method for optimizing the operational strategies of existing GSHPs while highlighting the value of process-accelerated optimization using numerical simulation-based surrogate models.