A hierarchical reinforcement learning-based vehicle-to-grid dispatch architecture for car parks: Integrating proximal policy optimisation and a large language model within a dual-agent framework

2026-06-27

PRE

Jinjia Cao

Xu Xu *

Weitao Yao

Fei Xue

龙超 (Chao Long)

DOI：10.1016/j.seta.2026.105158

Original

Original request for help

Save

Abstract

En 中文

• A fine-tuned LLM is integrated as a semantic reasoning module to adaptively reweight multi-objective rewards in real time. • A hierarchical dual-agent framework combines PPO-based decision making with convex optimisation-based execution for stable V2G scheduling. • The LLM-driven reweighting mechanism accelerates convergence and enhances adaptability under non-stationary grid conditions. • Natural-language reasoning improves the transparency and interpretability of reward adaptation and policy evolution.

Keywords:

Vehicle-to-grid

Deep reinforcement learning

Large language model

Battery swap station

Hierarchical

A hierarchical reinforcement learning-based vehicle-to-grid dispatch architecture for car parks: Integrating proximal policy optimisation and a large language model within a dual-agent framework

Abstract

Journal

Organization

Cited Papers

Citing Papers