Cyber-Physical Systems (CPS) and Internet of Things (IoT) applications are increasingly collaborating with LLM agents. A case study showcases the use of LLMs to simulate human preferences in a dynamic HVAC system optimization scenario.

Study Highlights:

• Demonstrated the use of LLM agents to represent various human demographics in a shopping mall environment.
• Developed a reinforcement learning algorithm, AitL-RL, that uses LLMs as a dynamic simulator for balancing energy and comfort.
• Revealed LLMs’ ability to imitate complex human behaviors and improve upon traditional set point control policies.
• Shared the project’s code on GitHub for community development.

Significance:

The research bridges the gap between advanced Foundation Models like LLMs and CPS-IoT applications, enhancing adaptability and efficiency. It suggests that:

• LLMs can drive more sustainable and user-friendly environment control solutions.
• There is potential for further exploration in LLM-integrated systems for real-time feedback and optimization.