Contextualized AI agents reveal the cognitive drivers of energy consumption

This repository contains the source code for the paper "Contextualized AI agents reveal the cognitive drivers of energy consumption" by Tao Wang and Yu Qian Ang, National University of Singapore.

Overview

This project introduces an in-silico laboratory to study the impact of human behavior on building energy consumption. It uses Large Language Model (LLM)-powered agents as digital human subjects within physically-accurate EnergyPlus simulations. The framework allows for the direct observation of agent reasoning, revealing the cognitive trade-offs between immediate comfort and future costs.

Repository Structure

• dynamic_ep_llm_multi.py: The main script for running the multi-agent EnergyPlus simulations.
• helper.py: Contains utility functions for the main simulation script.
• behavior_analysis.py: The main script for running the Behaviour Rationality Score analysis, together wit a report of the analysis results.
• comfort_cost_analysis.py: The main script for running the Comfort-Cost Trade-off analysis, quantifying the intertemporal discounting analysis.
• monthly_persona_viz.py: The main script for running the Monthly Persona analysis, visualizing the monthly patterns.

Citation

If you use this code in your research, please cite our paper:

Tao Wang, Yu Qian Ang. "Contextualized AI agents reveal the cognitive drivers of energy consumption." [TBC].