| 引用本文: | 李锦辉,吴毓峰,余 涛,潘振宁.数据孤岛下基于联邦学习的用户电价响应刻画及其应用[J].电力系统保护与控制,2024,52(6):164-176.[点击复制] |
| LI Jinhui,WU Yufeng,YU Tao,PAN Zhenning.Characterization of user price response behavior and its application based onfederated learning considering a data island[J].Power System Protection and Control,2024,52(6):164-176[点击复制] |
|
| 摘要: |
| 电价型需求响应离不开对用户价格响应的精准刻画，然而用户对价格的响应大多发生在与聚合商的交互中。并且出于隐私保护需求，这些交互数据往往不被公开，呈现为数据孤岛。针对现阶段用户数据隐私需求和电网调度需求相互冲突的问题，提出了基于联邦学习的用户电价响应行为刻画及其应用方法。首先，构建基于联邦学习的用户电价响应行为刻画的分布式交互框架，将原始数据信息交互转变为特征信息交互。然后，利用差分隐私-联邦近端算法实现不同聚合商电价响应模型的参数聚合，获得区域用户电价响应模型。最后，提出嵌入响应模型的配电网优化运行应用方法，利用改进的策略近端优化算法求解系统实时电价和储能出力。算例表明，所提方法在保障用户用能信息隐私下，能准确刻画区域用户电价响应行为，并改善配电网综合效益。 |
| 关键词: 电价型需求响应 用户隐私 联邦学习 强化学习 |
| DOI：10.19783/j.cnki.pspc.231013 |
| 投稿时间：2023-08-06修订日期：2023-11-14 |
| 基金项目:国家自然科学基金项目资助(52207105)；国家自然科学基金委员会-国家电网公司智能电网联合基金项目资助(U2066212)；中国博士后科学基金项目资助(2022M721184) |
|
| Characterization of user price response behavior and its application based onfederated learning considering a data island |
| LI Jinhui1,WU Yufeng1,YU Tao1,2,PAN Zhenning1 |
| (1. School of Electric Power, South China University of Technology, Guangzhou 510640, China; 2. Guangdong Provincial
Key Laboratory of Intelligent Measurement and Advanced Metering of Power Grid, Guangzhou 510640, China) |
| Abstract: |
| Price-based demand response cannot live without accurate characterization of user response, but these data mostly exist in the interaction between the demand response aggregator and users. Because of the imperative of privacy protection, the data often remains confidential, manifesting as isolated data islands. In response to the current problem where user data privacy requirements clash with the demands of grid scheduling, this paper proposes a method for characterizing user electricity price response behavior based on federated learning and its application. Initially, a distributed interactive framework for characterizing user electricity price response behavior through federated learning is established, transforming raw data information into feature-based interactions. Subsequently, a differential privacy-federated proximal algorithm is employed to aggregate parameters from various utility providers’ electricity price response models, resulting in a regional user electricity price response model. Finally, an application method for optimizing the operation of the distribution network is presented by embedding the response model. An improved strategy proximal optimization algorithm is used to determine real-time electricity prices and energy storage output. Case studies confirm that the proposed approach accurately characterizes regional user electricity price response behavior while preserving the privacy of user energy consumption information and enhancing the overall efficacy of the distribution network. |
| Key words: price-based demand response users’ privacy federated learning reinforcement learning |
