Artificial intelligence for shallow geothermal systems: A review of ground heat exchanger performance modeling

Shallow geothermal systems are increasingly recognized as a reliable and low-carbon solution for space heating and cooling, with ground heat exchangers (GHEs) serving as the core component for thermal energy exchange between buildings and the subsurface. The performance of GHEs, across boreholes, horizontal trenches, and energy geostructure configurations, is governed by complex, nonlinear interactions involving soil thermal properties, groundwater flow, material configurations, and system design parameters. Conventional modeling techniques often fall short in capturing these nonlinear and site-specific behaviors. In this context, artificial intelligence (AI) has emerged as a powerful tool for improving predictive accuracy, system optimization, and real-time decision-making in geothermal applications. This review presents a systematic assessment of AI applications in shallow GHE technologies, aiming to bridge the gap between geothermal engineering and modern data-driven modeling approaches. A total of 59 peer-reviewed studies were analyzed and classified according to GHE type, main application domain, and AI methodology. The review also includes AI-assisted interpretation of thermal response tests, which play a critical role in characterizing ground thermal properties prior to GHE system design. The analysis reveals two overarching patterns: (i) research efforts are heavily concentrated on borehole thermal-performance prediction, whereas thermomechanical behavior in energy geostructures and system-level operational optimization remain comparatively underexplored; and (ii) modern AI techniques such as physics-informed learning, uncertainty-aware models, reinforcement learning, and digital-twin concepts are rarely adopted, highlighting a clear methodological gap relative to broader AI advances. These findings outline emerging opportunities for next-generation AI frameworks that are physics-aware, uncertainty-quantified, and operationally adaptive.