LLM-BAS: Large Language Model-Based Battery Health-Aware Scheduling in Building Energy Systems

Battery energy storage systems are increasingly used in smart buildings, yet existing approaches often fail to integrate real-time battery health into energy scheduling. Most current methods either require extensive training data or lack interpretability and adaptability to battery degradation. This pa-per proposes LLM-BAS, Large Language Model-Based Battery Health-Aware Scheduling in Building Energy Systems that lever-ages Large Language Models (LLMs) to connect battery health diagnostics with adaptive control. The system consists of two agents: one for estimating state-of-health and degradation causes from telemetry data, and another for generating user-aligned charging policies based on diagnostic feedback and external context. A representative case study shows that the framework produces interpretable State of Health (SOH) assessments and conservative scheduling strategies aligned with lifespan extension. These findings demonstrate the feasibility of using LLMs as semantic controllers in energy-aware building systems.