AutoControl: An end-to-end fully automated workflow for control design of building energy systems

Developing an efficient and effective control system is critical to enhance built environment quality, to reduce building energy consumption and to achieve the carbon neutrality goal. However, this process is expertise demanding, time consuming, and error-prone, particularly for large-scale buildings with complicated building energy systems. To address this challenge, this study proposes AutoControl, an end-to-end fully automated workflow based on Large Language Models (LLMs). AutoControl incorporates two LLM-based agents that utilize Brick models as inputs to automatically design and generate Proportional–Integral (PI) controller codes. The first LLM agent, a semantic interpreter, extracts and translates configuration details from the Brick model into a comprehensive description of the building energy system and related control specifications. The second LLM agent, a control expert agent, generates the PI controller code based on the interpreted configurations. Finally, Particle Swarm Optimization (PSO) is employed to fine-tune the parameters of the PI controller. Experiments were conducted on three test cases with diverse HVAC system configurations using the BOPTest virtual testbed. AutoControl achieved an average Mean Absolute Error (MAE) in temperature of 0.323 ∘C and an average Root Mean Square Error (RMSE) in temperature of 0.766 ∘C during a week-period, demonstrating robust temperature control performance and strong generalization capabilities. These results highlight the potential of using LLMs for automatic development of building controllers.