Integrating Expert and Physics Knowledge for Modeling Heat Load in District Heating Systems
Francisco Souza, Thom Badings, Geert Postma, Jeroen Jansen
TL;DR
HELIOS tackles heat-load forecasting in district heating systems by fusing physics-based heat-balance modeling with expert knowledge via contextual mixture of experts. The model decomposes the load into $Q(k)=Q_{ ext{space}}(k)+Q_{ ext{hot-water}}(k)+Q_{ ext{loss}}(k)$ and learns parameters through MAP estimation that incorporate priors and context-driven gates. It provides interpretable components—space heating, hot-water use, and piping losses—while achieving state-of-the-art predictive accuracy on a Dutch DHS case study, outperforming static and dynamic baselines with strong short- to mid-term performance and clear explainability. The approach demonstrates the practical impact of integrating physics with expert knowledge for transparent, reliable energy management in DHS and beyond.
Abstract
New residential neighborhoods are often supplied with heat via district heating systems (DHS). Improving the energy efficiency of a DHS is critical for increasing sustainability and satisfying user requirements. In this paper, we present HELIOS, a dedicated artificial intelligence (AI) model designed specifically for modeling the heat load in DHS. HELIOS leverages a combination of established physical principles and expert knowledge, resulting in superior performance compared to existing state-of-the-art models. HELIOS is explainable, enabling enhanced accountability and traceability in its predictions. We evaluate HELIOS against ten state-of-the-art data-driven models in modeling the heat load in a DHS case study in the Netherlands. HELIOS emerges as the top-performing model while maintaining complete accountability. The applications of HELIOS extend beyond the present case study, potentially supporting the adoption of AI by DHS and contributing to sustainable energy management on a larger scale.