HB-PINNs: A Heat-Balance Physics-Informed Framework for Building Energy Loading Forecasting with Limited Data

In the context of global decarbonization, buildings have received increasing attention due to their significant share in total energy consumption. Accurate prediction of building energy demand and indoor temperature forms the basis for effective control, demand response, and integrated energy management. Physics-informed neural networks (PINNs) offer a promising way to combine physical knowledge with data-driven learning. However, most existing work in the building domain focuses on subsystem-level tasks or control applications, with limited efforts directed toward whole-building performance prediction. In the limited studies that do consider whole-building applications, simplified resistance capacitance (RC) models are almost exclusively used as the physical constraint, since they are easy to formulate and computationally efficient. However, conventional RC models rely on simplified thermal representations and cannot explicitly represent some latent thermal processes, which may limit their prediction accuracy under varying operating conditions. In this study, a new method called heat-balance PINNs (HB-PINNs) is introduced. It directly uses the formula of heat balance to find out the latent physical quantities that can’t be measured easily. In this way, the model can reconstruct a complete physical representation, so that the heating demand can be calculated more accurately. Under identical input conditions, HB-PINNs is evaluated against an RC-based PINN. The evaluation is carried out across multiple test sets, various training data ratios, and different prediction horizons. The proposed HB-PINNs achieves an average ????2 of 94.99% across all scenarios and training ratios, compared with 88.69% for RC-PINNs. The results show that HB-PINNs is more accurate, and it has stronger adaptability under different operating conditions. These findings show that HB-PINNs provides an extensible and physical solution for building energy consumption modelling.