Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimizing Electric Vehicles Charging using Large Language Models and Graph Neural Networks

Stavros Orfanoudakis, Peter Palensky, Pedro P. Vergara

TL;DR

The paper tackles the challenge of optimizing EV charging to maintain grid stability as EV adoption grows, where traditional optimization and RL struggle with high dimensionality and dynamics. It proposes a hybrid approach that uses Graph Neural Networks to model relational structure in the charging network and Large Language Models via a Decision Transformer to learn long-horizon charging policies from offline data $\mathcal{D}=\{(s,a,r)\}$ by maximizing $\mathbb{E}\left[\sum_{t=0}^{\infty}\gamma^t R(s_t,a_t)\right]$. The contributions include (i) a graph-based embedding for charging infrastructure, (ii) integration of LLMs for EV charging decision-making, and (iii) empirical results showing superior performance to baselines including CAFAP, BaU, PPO, and even offline Optimal in a 50-station scenario. The work demonstrates potential for scalable, real-time V2G optimization and suggests pathways to extend LLM-GNN hybrids to other grid decision problems.

Abstract

Maintaining grid stability amid widespread electric vehicle (EV) adoption is vital for sustainable transportation. Traditional optimization methods and Reinforcement Learning (RL) approaches often struggle with the high dimensionality and dynamic nature of real-time EV charging, leading to sub-optimal solutions. To address these challenges, this study demonstrates that combining Large Language Models (LLMs), for sequence modeling, with Graph Neural Networks (GNNs), for relational information extraction, not only outperforms conventional EV smart charging methods, but also paves the way for entirely new research directions and innovative solutions.

Optimizing Electric Vehicles Charging using Large Language Models and Graph Neural Networks

TL;DR

The paper tackles the challenge of optimizing EV charging to maintain grid stability as EV adoption grows, where traditional optimization and RL struggle with high dimensionality and dynamics. It proposes a hybrid approach that uses Graph Neural Networks to model relational structure in the charging network and Large Language Models via a Decision Transformer to learn long-horizon charging policies from offline data by maximizing . The contributions include (i) a graph-based embedding for charging infrastructure, (ii) integration of LLMs for EV charging decision-making, and (iii) empirical results showing superior performance to baselines including CAFAP, BaU, PPO, and even offline Optimal in a 50-station scenario. The work demonstrates potential for scalable, real-time V2G optimization and suggests pathways to extend LLM-GNN hybrids to other grid decision problems.

Abstract

Maintaining grid stability amid widespread electric vehicle (EV) adoption is vital for sustainable transportation. Traditional optimization methods and Reinforcement Learning (RL) approaches often struggle with the high dimensionality and dynamic nature of real-time EV charging, leading to sub-optimal solutions. To address these challenges, this study demonstrates that combining Large Language Models (LLMs), for sequence modeling, with Graph Neural Networks (GNNs), for relational information extraction, not only outperforms conventional EV smart charging methods, but also paves the way for entirely new research directions and innovative solutions.

Paper Structure

This paper contains 8 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. EV Smart Charging
  3. LLMs for Smart Charging
  4. Data Gathering for Offline RL
  5. GNN Embeddings
  6. Learning with LLMs
  7. Results
  8. Conclusions

Figures (2)

  • Figure 1: Overview of the proposed learning method. Trajectories are first generated by interacting with the environment using optimal and random policies. The resulting state and action vectors are then transformed into graph representations to efficiently capture relational dependencies and support dynamic environments. Finally, an offline RL method, powered by an LLM, is trained to optimize charging strategies and maximize future rewards.
  • Figure 2: Visual comparison of a use case with 50 chargers showing: a) an individual charger's charging profile w.r.t. the connected EVs, and b) the aggregated power following the grid operator's power limit.