Model predictive control for intelligent PID self-tuning in HVAC systems achieves adaptable energy savings in ultra-low energy buildings

In ultra-low energy buildings, maintaining the HVAC efficiency under dynamic operating conditions remains a critical challenge, particularly for conventional proportional-integral-derivative (PID) controllers with fixed parameters, which lack the adaptability to handle system nonlinearities and variable meteorological loads. To address this, this study develops and comparatively evaluates two model predictive control (MPC)-oriented strategies, namely neural network‑based and fuzzy‑logic‑based (denoted as NN‑MPC and FLC‑MPC for brevity), for the real-time PID self-tuning in a ground-source heat pump system in an ultra-low energy building. The core innovation lies in the FLC-MPC’s rule-based architecture, which excels in adapting to system nonlinearities and operational uncertainties over conventional data-driven approaches. Simulation results obtained from a calibrated TRNSYS model reveal that, the main contribution is a direct and quantitative demonstration of this adaptability, for which the NN-MPC achieves a mean saving of 1.72% over the two test days, whereas the FLC-MPC delivers a greater average saving of 5.55% with a peak of 16.46% on the tested day of August 10, while maintaining the indoor temperature at the fixed comfort setpoints specified in the simulation. Beyond this, the proposed data-driven MPC framework offers significant application potential, by being integrated into existing Building Management Systems, enabling the data-informed PID parameter tuning without hardware replacement. This facilitates the self-adaptive HVAC control that effectively balances energy efficiency and thermal comfort under diverse meteorological conditions, suggesting its potential applicability for next-generation low-energy building operations.