【Applied Energy】基于联邦学习的关注用户隐私保护的激励需求响应优化方法

AEii国际应用能源 · 公众号 · · 2024-01-22 18:30

正文

原文信息：

Incentive-based demand response optimization method based on federated learning with a focus on user privacy protection

原文链接：

https://www.sciencedirect.com/science/article/pii/S0306261923019347

Highlights

(1) Users’ influence models can accurately reflect historical power consumption characteristics.

(2) Privacy and response incentive mechanisms collectively guide the optimal scheduling of demand response.

(3) The federated learning-based optimization method achieves economic Nash equilibrium while upholding users’ privacy.

(4) Optimal strategies for the distribution network manager vary based on different privacy requirements.

摘要

考虑到需求响应(DR)背景下众多灵活能源用户的灵活容量和隐私需求，本文建立了考虑隐私预算的用户参与DR能力影响模型(IM)。利用所设计的Stackelberg博弈机制，从IM描述的不同特征的用户中实现DR响应者的最优选择，提出了一种基于联邦学习(FL)的优化方法，该方法以差分隐私(DP)作为DR经济最优调度的数据传输机制。基于FL的优化方法控制参与用户数量和对隐私泄露风险进行财务补偿的思想是拥有私人数据的用户参与DR的保证。在案例研究中，本文提出的优化方法与蛾火优化算法的性能进行了比较，并讨论了前者在不同特征的用户群体中选择的指导价值。结果表明，本文提出的方法具有良好的经济效益和普遍适用性。

Abstr act

Considering the flexible capacity and privacy needs of numerous flexible energy users in the context of demand response (DR), this study establishes an influence model (IM) to describe the DR participation capabilities of users considering privacy budget. Using the designed Stackelberg game mechanism that can achieve optimal selection for DR responders from users with different characteristics described by IM, a federated learning (FL)- based optimization method that uses differential privacy (DP) as the data transmission mechanism for DR economic optimal dispatch is proposed. Ideas of controlling the number of participating users and financially compensating for the privacy leakage risk by the FL-based optimization method are the guarantees for users with private data to participate in DR. The performance of the proposed optimization method is also compared with that of the Moth-flame optimization algorithm in a case study, and the guiding value of the former in selecting among user groups with different characteristics is then discussed. Results show that the proposed method exhibits good economic benefits and universal applicability.

Keywords

Demand response

Influence model

Stackelberg game

Differential privacy

FL-based optimization method

Graphics