ThermoLLM: Thermodynamics-Aware HVAC Control with Spatial-Semantic Knowledge Graph

The paper introduces ThermoLLM, a thermodynamics-aware HVAC control framework that leverages a physics-informed spatial knowledge graph for managing a five-zone EnergyPlus building simulation. By integrating building semantics with recent interaction history, the model enhances decision-making regarding thermal dynamics and zone coupling, outperforming standard control baselines and LLM-based alternatives in energy-comfort trade-offs and PMV violations. This framework is significant for practitioners as it demonstrates a novel approach to HVAC control that incorporates structured spatial reasoning, potentially leading to more efficient and responsive building management systems.