An integrated data-driven modeling and dynamic optimization framework for ground source heat pump systems in an operational nearly zero-energy building

The energy optimization of ground-source heat pump (GSHP) systems in nearly zero-energy buildings (NZEBs) remains challenging due to the disconnect between subsystem-level modeling and system-wide optimization, as well as the simulation-reality gap. To bridge these gaps, this study proposes and validates an integrated data-driven framework that synergizes multi-scale predictive modeling with an enhanced real-time optimization algorithm. The core contribution of this work lies in the closed-loop integration of dynamic indoor thermal response, equipment power, and building load into a single operational optimization pipeline, validated through live field-deployment. First, a competitive co-optimization methodology was employed to develop high-accuracy black-box models for building load and equipment energy consumption using Building Energy Management System (BEMS) data and calibrated TRNSYS simulations of a real NZEB in Beijing. Second, a gray-box resistance-capacitance (RC) thermal model, calibrated via a genetic algorithm, was established to provide high-fidelity indoor temperature predictions. Finally, a composite adaptive Particle Swarm Optimization (PSO) algorithm was designed to dynamically optimize system operation. The results demonstrate that the optimal predictive models achieved a validation R2 of 0.87 for load and a mean absolute percentage error (MAPE) of 5.18% for system power, while the RC model attained an R2 of 0.97 for indoor temperature. The enhanced PSO algorithm accelerated convergence by a factor of 4.67 compared to the standard version. When deployed in the real building, the framework reduced the total heating energy consumption by 12.4% under winter conditions while maintaining thermal comfort, with component-level savings of 15.3% for the GSHPs and 8.1% for the pumping subsystem. This study provides a practical, field-proven paradigm for bridging multi-scale building energy simulations with real-time, physical self-adaptation in NZEBs.