Skip to main navigation Skip to search Skip to main content

A Deep Reinforcement Learning Approach for Multi-Agent Coordinated Optimization in V2G Scenarios

  • Dongjie Sun*
  • , Zhongbao Wei
  • *Corresponding author for this work
  • Beijing Institute of Technology

Research output: Contribution to journal › Conference article › peer-review

Abstract

The rapid proliferation of electric vehicles (EVs) introduces significant challenges to traditional distribution networks, including increased peak-to-valley differences, aggravated voltage fluctuations, and higher line losses. To address these issues, Vehicle-to-Grid technology, coupled with advanced dispatch mechanisms, is crucial. This work proposes an innovative multi-agent deep reinforcement learning (MADDPG)-based framework to achieve collaborative optimal dispatch of EV charging/discharging within distribution networks. The framework effectively balances the complex, multi-level coupled interests among distribution network operators, EV aggregators (EVAs), and EV users. By leveraging real-world grid load, electricity price, and EV travel data, the proposed MADDPG model demonstrates superior performance compared to uncoordinated strategies. Specifically, it achieves a 19.3% reduction in the distribution network’s peak-to-valley difference and a 20.8% decrease in load fluctuation, significantly enhancing grid stability. Concurrently, EVAs profits are boosted by 18.5%, while EV user satisfaction improves by 3.4% through optimized charging costs and discharging revenues.

Original languageEnglish
Article number012018
JournalJournal of Physics: Conference Series
Volume3214
Issue number1
DOIs
Publication statusPublished - 2026
Externally publishedYes
Event8th International Conference on Energy Storage and Intelligent Vehicles, ICEIV 2025 - , Singapore
Duration: 5 Dec 2025 → 7 Dec 2025

Fingerprint

Dive into the research topics of 'A Deep Reinforcement Learning Approach for Multi-Agent Coordinated Optimization in V2G Scenarios'. Together they form a unique fingerprint.

Cite this