What question did this study set out to answer?

April 18, 2026Advances in Applied Energy0 citationsOpen Access

Coordinated Interaction of Electric Vehicles, Charging Stations, and the Power Grid Using Multi-Agent Reinforcement Learning

YLYuchen LiTsinghua University HZH ZhaoTsinghua University ZSZhixuan SongTsinghua University

Key Points

The research aims to enhance the coordination between electric vehicles, charging stations, and the power grid using a multi-agent reinforcement learning framework.
Proposed a transformer-enhanced multi-agent actor-critic framework for coordination.
Integrated factors like renewable generation, EV storage, battery degradation, and dynamic tariffs.
Utilized a centralized-training–decentralized-execution approach for cooperative decision-making.
Charging stations increased profits by 33.5% through dynamic pricing.
Electric vehicle users reduced costs by 4.3% via energy storage participation.
Achieved a 40.7% redistribution of renewable energy across spatial and temporal dimensions.

Abstract

• Proposes a transformer-enhanced MARL framework for EV–Station–Grid coordination. • Integrates renewable generation, EV storage, battery degradation, and dynamic tariffs. • Enables EVs to redistribute 40.7% of renewable energy across time and space. • Dynamic pricing boosts station profit by 33.5% and cuts user costs by 4.3%. • Proposed framework unlocks EVs as flexible storage for grid, operators, and users. While the rapid growth of electric vehicles poses significant challenges to grid stability through increased peak demand, their potential as distributed energy storage to mitigate renewable intermittency and provide grid flexibility could bring substantial benefits. Existing studies have addressed either charging station pricing or vehicle scheduling independently. However, the cooperative game-theoretic co-optimization of dynamic pricing and bidirectional charging scheduling of electric vehicles within a vehicle-station-grid ecosystem remains underexplored. This study bridges this gap by proposing a Transformer-enhanced multi-agent actor-critic framework with a centralized-training–decentralized-execution paradigm, enabling cooperative decision-making where the station optimizes pricing strategies and vehicles schedule bidirectional charging in response to renewable generation and electricity prices. Battery aging costs are explicitly embedded in rewards to prevent degradation-intensive cycling. Simulation results demonstrate that the charging station can achieve a 33.5% increase in sustained profits through automated dynamic pricing, while electric vehicle users reduce electricity costs by 4.3% through participating in energy storage. From a power system perspective, the framework efficiently redistributes 40.7% of renewable energy across both spatial and temporal dimensions. These findings demonstrate that coordinated multi-agent reinforcement learning unlocks economic potential for both stations and users while enhancing renewable utilization and reducing grid pressure.

AIに質問

Bookmark

View Full Paper