LLM-Powered Energy Optimizer
The LLM-Powered Energy Optimizer integrates Large Language Models (LLMs) with the CityLearn multi-building energy environment to achieve adaptive energy coordination and optimization.
By combining the reasoning ability of LLMs with reinforcement-based control agents, this framework demonstrates how language-driven guidance can enhance decision-making and interpretability in smart energy systems.
System Overview
The project leverages CityLearn, an open-source urban energy simulator, where multiple buildings interact with shared energy resources such as batteries, HVAC systems, and renewable units.
The LLM acts as a high-level advisor, generating structured reasoning steps, policy explanations, and optimization prompts to assist the RL controller in selecting efficient energy actions.
Core Components
- CityLearn Environment: Multi-building coordination for energy storage and HVAC scheduling
- LLM Agent Layer: Generates task explanations, reasoning chains, and adaptive control advice
- Reinforcement Learning Backbone: PPO and SAC algorithms for optimizing power and comfort metrics
- Physics-Informed Feedback: Embeds voltage, power, and thermal constraints into decision flow
Objectives
- Reduce total energy consumption and carbon footprint
- Improve system interpretability through LLM reasoning chains
- Enable human-in-the-loop adjustments using natural-language instructions
Research Impact
This work bridges the gap between natural-language intelligence and energy optimization, enabling explainable, sustainable, and LLM-guided adaptive control in modern smart cities.
