An LLM-assisted decision-making framework of rule-based control strategy for photovoltaic driven air conditioning systems

Photovoltaic driven air conditioning (PVAC) systems provide significant opportunities for sustainable building energy management, but optimizing their operational parameters remains challenging. Current control methods often lack flexibility in adapting to dynamic environmental conditions and occupancy requirements. This study introduces a novel optimization framework integrating a large language model (LLM) with a rule-based adaptive control strategy for PVAC systems. A comprehensive simulation model, coupling detailed building thermal dynamics and air conditioning behavior, is developed and experimentally validated. The LLM is requested to determine optimal operational parameters. The LLM is systematically evaluated across varying scenarios, including initial settings for temperature setpoints, comfort ranges, hysteresis bands, and communication intervals. Results indicated that the LLM consistently converged on optimal control configurations, notably temperature setpoints around 24–25 °C, comfort ranges of approximately 1–2 °C, and hysteresis bands of about 0.3–0.5 °C, achieving high thermal comfort satisfaction and substantial energy cost savings. Furthermore, the model demonstrated excellent repeatability and operational stability across diverse climatic conditions. This research confirms the viability of integrating LLM-assisted adaptive control into PVAC systems. The LLM provides a promising opportunity for enhanced building energy efficiency and occupant comfort in practical applications.