CarbonGPT: Meta Causal Graph-Enhanced Large Language Models for Carbon Emission Forecasting in Large-Scale Power Distribution Networks

TL;DR

This paper proposes CarbonGPT, a meta causal graph-enhanced LLM for accurate carbon emission forecasting in power distribution networks with high EV and RES penetration.

Problem

Predicting dynamic carbon flow in power distribution networks is challenging due to the electrification of transport and renewable energy volatility. Existing adaptive graph and LLM methods can suffer from amplified spurious correlations, limiting their learning and reasoning.

Approach

CarbonGPT uses a causal encoder to uncover true cross-variable causal relationships. It incorporates a meta causal graph dictionary and lightweight alignment to improve the LLM's comprehension of carbon feature representations in EV and RES integration scenarios.

Results & Contribution

Extensive simulations in PDNs under large-scale EV and RES integration show that CarbonGPT achieves state-of-the-art performance in both prediction accuracy and effectiveness. The code is available at https://github.com/lishijie15/CarbonGPT.