| 引用本文: | 段玉戈,王小君,刘曌,司方远,许寅,和敬涵,葛路琨,王智爽.基于先验知识修正的社区充电站在线实时优化调度[J].电力自动化设备,2026,46(2):137-147. |
| DUAN Yuge,WANG Xiaojun,LIU Zhao,SI Fangyuan,XU Yin,HE Jinghan,GE Lukun,WANG Zhishuang.Online real-time optimal scheduling of community charging stations based on prior knowledge correction[J].Electric Power Automation Equipment,2026,46(2):137-147. |
|
| 摘要: |
| 传统算法在求解社区充电站的在线实时优化调度过程中因用户需求多元、负荷接入随机、模型计算复杂度高等问题而难以同时兼顾精度和效率。针对考虑充电信息与能量复杂在线交互的社区充电站运行决策问题，提出一种基于先验知识修正的社区充电站在线实时优化调度策略。考虑电动汽车用户的充电紧迫度，刻画用户的时序充电需求，建立以最小化充电站运营成本为优化目标的协调调度模型；采用模型机理驱动的运筹学优化方法生成先验知识，并将其嵌入智能体以提升探索效率；基于用户时序个性化充电数据，提出基于先验知识修正的深度确定性策略梯度算法，从而获得最优连续动态调度决策。以2座社区充电站的实际运行数据为算例进行仿真测试，结果表明本文所提调度策略使2座社区充电站的月度运行成本分别平均降低了4.664 %和5.366 %。通过分析比较不同算法在线优化后2座充电站的负荷指标和算法时间效益，进一步验证了所提调度策略在降低社区充电站运行负荷峰谷差、提高算法求解效率方面的可行性和高效性。 |
| 关键词: 电动汽车 社区充电站 先验知识 深度强化学习 在线实时调度 |
| DOI：10.16081/j.epae.202507025 |
| 分类号: |
| 基金项目:智能电网国家科技重大专项(2030)资助项目(2024ZD0800800)；国家电网有限公司总部科技项目资助(低碳高可靠城市配电系统示范工程)(SGTJDK00DWJS2400298) |
|
| Online real-time optimal scheduling of community charging stations based on prior knowledge correction |
|
DUAN Yuge1, WANG Xiaojun1, LIU Zhao1, SI Fangyuan1, XU Yin1, HE Jinghan1, GE Lukun2, WANG Zhishuang2
|
|
1.School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China;2.State Grid Tianjin Electric Power Research Institute, Tianjin 300392, China
|
| Abstract: |
| In the process of solving the online real-time optimal scheduling of community charging stations, traditional algorithms face inherent challenges in balancing accuracy and efficiency simultaneously because of the problems such as diverse user demands, random load access, high computational complexity of models, and so on. Aiming at the operation decision-making problem of community charging stations considering the complex online interaction of charging information and energy, an online real-time optimal scheduling strategy of community charging stations based on prior knowledge correction is proposed. Considering the charging urgency of electric vehicle users, the users’ time-series charging demand is characterized and a coordinated scheduling model is established with the objective of minimizing the operation cost of charging stations. The prior knowledge is generated by using model mechanism-driven operations research method and then embedded within agents to improve exploration efficiency. Based on the users’ time-series personalized charging data, a prior knowledge correction-based deep deterministic policy gradient algorithm is proposed to obtain the optimal continuous dynamic scheduling decision-making. Taking the actual operation data of two community charging stations as a case for simulation testing, the results show that the proposed scheduling strategy reduces the average monthly operation cost of the two community charging stations by 4.664 % and 5.366 % respectively. By analyzing and comparing the load indicators and algorithm time efficiency of two charging stations after online optimization with different algorithms, the feasibility and efficiency of the proposed scheduling strategy in reducing the peak-valley difference of community charging station operation load and improving the algorithm solution efficiency are further verified. |
| Key words: electric vehicles community charging station prior knowledge deep reinforcement learning online real-time scheduling |
