一种面向大规模电动汽车集群的隐私保护安全调度方法

Privacy Preserving Secure Scheduling Method for Large Scale Electric Vehicle Clusters

摘要

大规模电动汽车集群参与的电动汽车-虚拟电厂(EV-VPP)无需额外投资即可提升对聚合分布式能源参与电网调频的能力．然而，其聚合调控依赖于EV位置和电池荷电状态等隐私信息，易削弱车主参与积极性并降低集群调控潜力．针对EV集群聚合过程中的隐私保护问题，本文提出一种强化安全保护的横向联邦强化学习(SEFRL)调度方法．首先，构建了基于马尔可夫过程的EV-VPP优化调度模型．其次，通过本地差分隐私扰动单个EV-VPP的原始上传数据，实现针对上层调度中心的数据安全防护．然后，通过Kyber-AES轻量级后量子加密方法保护联邦聚合数据，有效阻止通过对联邦学习梯度参数的监听而推演出原始隐私数据，实现信道的安全防护，并实现安全性和计算效率之间的平衡．最后，在改进IEEE-33节点系统中构建包括EV集群在内多种分布式能源参与的EV-VPPs协同调度系统验证所提算法性能．仿真表明，面对Deep Leakage推断攻击，所提SEFRL方法可以稳定保持联邦学习梯度参数反推演隐私数据的误差在0.91以上，保证真实数据无法获知，从而实现针对上层调度中心的数据保护．此外，隐私保护能力的提升增强了EV集群参与调度的积极性，系统整体调频成本相较于EV无序充电降低了29.5%．

Abstract

The large-scale integration of electric vehicle(EV)clusters in virtual power plants(EV-VPPs)enhances the ability of these aggregated distributed energy resources to provide grid frequency regulation without requiring additional investment. However，controlling these systems relies on private information，such as EV locations and bat- tery state of charge，which may undermine EV owners’ willingness to participate and diminish the regulatory poten- tial of the clusters. To address these privacy concerns during EV cluster aggregation this study proposes a security enhanced federated reinforcement learning(SEFRL)scheduling method. First，an optimal EV-VPP scheduling model is constructed based on a Markov decision process. Second，local differential privacy is applied to perturb the raw data uploaded by individual EV-VPPs，thereby safeguarding data privacy from the perspective of the upper-level scheduling center. Then，a lightweight postquantum Kyber-AES hybrid encryption scheme is introduced to secure the transmission of aggregated federated data，thereby effectively preventing the inference of raw private information through eavesdropping on gradient parameters while balancing security with computational efficiency. Finally，the proposed method is validated using a coordinated scheduling system that integrates EV clusters with other distributed energy resources on a modified IEEE-33 bus system. The simulation results indicate that under deep leakage inference attacks，the SEFRL method maintains a stable error rate above 0.91 when back-calculating private data from federated learning gradients，thereby effectively protecting data from the upper-level scheduling center. Moreover， enhanced privacy protection increases EV-cluster participation in scheduling，resulting in a 29.5% reduction in overall system-frequency-regulation costs compared with uncoordinated charging strategies.

关键词

电动汽车集群 / 本地差分隐私 / 隐私保护 / 后量子加密

Key words

electric vehicle(EV) cluster / local differential privacy / privacy preservation / postquantum encryption

参考文献

[1] International Energy Agency(IEA). Global EV Outlook 2024[EB/OL]. https://www.iea.org/reports/global-ev-outlook-2024. 2025-04-05.

[2] 葛磊蛟, 李元良, 汪宇倩. 智能配电网态势感知实现效果综合评估模型[J]. 天津大学学报(自然科学与工程技术版), 2020, 53(11):1101-1111.

[3] Ge Leijiao, Li Yuanliang, Wang Yuqian. Comprehensive evaluation model for situational awareness effects of a smart distribution network[J]. Journal of Tianjin University(Science and Technology), 2020, 53(11) : 1101-1111(in Chinese).

[4] 王明深, 于汀, 穆云飞, 等. 电动汽车能效电厂价格响应模型[J]. 天津大学学报(自然科学与工程技术版), 2016, 49(12):1320-1329.

[5] Wang Mingshen, Yu Ting, Mu Yunfei, et al. A price response model for efficient power plant of electric vehicles[J]. Journal of Tianjin University(Science and Technology), 2016, 49(12):1320-1329(in Chinese).

[6] Ledro M, Calearo L, Zepter J M, et al. Influence of realistic EV fleet response with power and energy controllers in an EV-wind virtual power plant[J]. Sustainable Energy Grids & Networks, 2022, 31:100704.

[7] Ebrahimi M, Ebrahimi M, Shafie-Khah M, et al. EV observing distribution system management considering strategic VPPs and active & reactive power markets[J]. Applied Energy, 2024, 364:123152.

[8] Qiu R X, Liu X, Huang R, et al. Differential privacy EV charging data release based on variable window[J]. PeerJ Computer Science, 2021, 7:e481.

[9] Lu C B, Wu J M, Wu C Y. Privacy-preserving decentralized price coordination for EV charging stations[J]. Electric Power Systems Research, 2022, 212:108355.

[10] 郭静, 顾智敏, 朱道华, 等. 隐私保护的电动汽车充电行为安全预测方法[J]. 电讯技术, 2025, 65(7):1033-1041.

[11] Guo Jing, Gu Zhimin, Zhu Daohua, et al. A secure charging behaviour forecasting method with privacy protection[J]. Telecommunication Engineering, 2025, 65(7):1033-1041(in Chinese).

[12] Yu H, Zhang Y L, Qu J H, et al. A privacy-protected distributed operation method for flexible distribution networks with EV charging load clusters[J]. Energy, 2025, 327:136409.

[13] 李元诚, 胡柏吉, 黄戎. 基于匿名凭证与区块链的V2G网络电力交易隐私保护认证方案[J]. 通信学报, 2025, 46(5):145-158.

[14] Li Yuancheng, Hu Boji, Huang Rong. Privacy-preserving authentication scheme for electricity trading in V2G network using anonymous credential and blockchain[J]. Journal of Communications, 2025, 46(5): 145-158(in Chinese).

[15] 杨挺, 覃小兵, 冯相为, 等. 计及用户充电行为与隐私保护的联邦学习电动汽车短期充电负荷预测[J]. 高电压技术, 2024, 50(10):4512-4519.

[16] Yang Ting, Qin Xiaobing, Feng Xiangwei, et al. Short-term charging load prediction of federated learning electric vehicles after accounting for user charging behavior and privacy protection[J]. High Voltage Engineering, 2024, 50(10):4512-4519(in Chinese).

[17] Sharma A, Marchang N. A review on client-server attacks and defenses in federated learning[J]. Computers & Security, 2024, 140:103801.

[18] Csatár J, György P, Holczer T. Holistic attack methods against power systems using the IEC 60870-5-104 protocol[J]. Infocommunications Journal, 2023, 15(3): 42-53.

[19] Hu H S, Zhang X Y, Salcic Z, et al. Source inference attacks:Beyond membership inference attacks in federated learning[J]. IEEE Transactions on Dependable and Secure Computing, 2024, 21(4):3012-3029.

基金资助

国家重点研发计划资助项目(2022YFB2403900)

国家自然科学基金资助项目(62371338)