首页|期刊导航|福建师范大学学报(自然科学版)|光储充电站中基于斯塔克伯格博弈的电动汽车充电调度优化策略

光储充电站中基于斯塔克伯格博弈的电动汽车充电调度优化策略OA

A Stackelberg game-based optimization strategy for EV charging scheduling in photovoltaic storage charging stations

中文摘要英文摘要

针对光储充电站中电动汽车无序充电问题,提出一种基于斯塔克伯格博弈的多智能体深度强化学习调度策略,旨在通过优化充电调度与动态定价最大化充电站运营收益.将充电站与电动汽车用户间的双层互动建模为斯塔克伯格马尔可夫博弈,充电站作为领导者制定价格,用户作为跟随者响应价格,该模型贴近真实商业场景且能保护用户隐私.采用基于斯塔克伯格博弈的多智能体双延迟确定性策略梯度算法求解,该算法结合斯塔克伯格分布式框架与多智能体双延迟深度确定性策略梯度(multi-agent twin delayed deep deterministic policy gradient,MATD3)的稳定学习机制,并引入优先经验回放提升数据利用率与收敛速度.实验结果表明,与多智能体深度确定性策略梯度(multi-agent deep deterministic policy gradient,MAD-DPG)和 MATD3 算法相比,所提算法运营收益分别提升 20.9%和 32.2%;与精确建模 KKT 算法相比,综合收益达 90.9%,且决策时间不随电动汽车数量增加而大幅增长.算法学习到的策略能有效利用 ESS(en-ergy storage system)实现削峰填谷,平滑电网负荷,验证了其有效性与实用性.

To address the issue of uncoordinated electric vehicle charging in photo-voltaic-storage charging stations,this paper proposes a multi-agent deep reinforcement learning scheduling strategy based on a Stackelberg game,aiming to maximize the charging station's operat-ing revenue by optimizing charging scheduling and dynamic pricing.The bi-level interaction between the charging station and EV users is modeled as a Stackelberg Markov game,where the charging station acts as the leader in setting prices,and users act as followers in responding to them.This model closely reflects real-world business scenarios and helps protect user privacy.The proposed model is solved using a Stackelberg game-based multi-agent twin delayed deep deterministic policy gradient(MATD3)algorithm,which integrates the distributed Stackelberg framework with the sta-ble learning mechanism of MATD3 and introduces prioritized experience replay to enhance data utili-zation and convergence speed.Experimental results show that compared with MADDPG and MATD3,the proposed algorithm increases operating revenue by 20.9%and 32.2%,respectively.Compared with the KKT-based exact modeling algorithm,it achieves 90.9%of the comprehensive revenue,and its decision-making time does not increase substantially with the number of EVs.The learned strategy effectively utilizes the energy storage system for peak shaving and valley filling,thereby smoothing the grid load and demonstrating its effectiveness and practicality.

陈愉;林兴源;何凱;卢宇;林兵

福建师范大学物理与能源学院,福建 福州 350117澳门科技大学计算机科学与工程学院,澳门 999078福建师范大学物理与能源学院,福建 福州 350117福建师范大学物理与能源学院,福建 福州 350117||福建师范大学协和学院,福建 福州 350117福建师范大学物理与能源学院,福建 福州 350117||福建省网络计算与智能信息处理重点实验室,福建 福州 350116

信息技术与安全科学

电动汽车充电调度斯塔克伯格博弈多智能体深度强化学习

electric vehiclescharging schedulingStackelberg gamemulti-agent deep re-inforcement learning

《福建师范大学学报(自然科学版)》 2026 (5)

1-12,12

国家自然科学基金项目(62072108)福建省高校产学合作项目(2022H6024、2021H6026)福建省高校物理学学科联盟教学改革项目(FJPHYS-2022-B02)福建省促进海洋与渔业产业高质量发展专项资金(FJHYF-ZH-2023-02)福建省技术创新重点攻关及产业化项目(2024XQ004)福建省科技厅引导性项目(2023H0009)

10.12046/j.issn.1000-5277.2025100002

评论