基于纵向联邦学习的短期风电功率协同预测方法  

Collaborative Forecasting Method for Short-term Wind Power Based on Vertical Federated Learning

作　　者：赵寒亭 张耀[1] 霍巍 王建学[1] 吴峰 张衡 ZHAO Hanting;ZHANG Yao;HUO Wei;WANG Jianxue;WU Feng;ZHANG Heng(School of Electrical Engineering,Xi'an Jiaotong University,Xi'an 710049,China;Dahang Youneng Electrical Co.,Ltd.,Yangzhong 212211,China)

机构地区：[1]西安交通大学电气工程学院,陕西省西安市710049 [2]大航有能电气有限公司,江苏省扬中市212211

出　　处：《电力系统自动化》2023年第16期44-53,共10页Automation of Electric Power Systems

基　　金：国家自然科学基金资助项目(51907151);陕西省自然科学基础研究计划资助项目(2023-JC-QN-0391)。

摘　　要：由于风力资源具有时空相关性,使用邻近场站的相关数据可以提高待预测场站的预测精度。然而不同场站通常分属不同发电集团,由于商业竞争和数据安全考量,彼此难以获得对方的隐私数据。针对上述问题,首先,提出了基于改进k近邻算法的岭回归预测模型;然后,在纵向联邦学习的机制下,采用同步梯度下降算法对所提预测模型进行迭代求解;利用梯度向量可拆分计算的特点,推导了风电预测模型的分布式训练过程和分布式预测过程,将原本的大规模预测问题分解为大量的小规模子问题,且每个子问题由相应的风电场站在本地进行计算。在保证各参与方数据隐私安全的基础上,可以有效利用邻近场站的数据信息,从而提高风电功率预测精度。最后,以实际算例验证了所提方法的有效性。Because of the spatio-temporal correlation of wind resources,the forecasting accuracy of the forecasted wind farm can be improved by using the relevant data of adjacent wind farms.However,different wind farms are often owned by different power generation groups,which do not have access to private data of each other due to commercial competition and data security concerns.To solve these problems,first,a ridge regression forecasting model based on the improved k-nearest neighbor algorithm is proposed in this paper.Then,under the mechanism of vertical federated learning,the synchronous gradient descent algorithm is used to solve the proposed forecasting model iteratively.The distributed training process and the distributed forecasting process of wind power forecasting model are derived by utilizing the separable characteristic of gradient vector calculation.The original largescale forecasting problem is decomposed into a large number of small-scale subproblems,and each subproblem is calculated locally by the corresponding wind farm.On the basis of ensuring the data privacy and security of all participants,the data information of adjacent wind farms can be effectively used to improve the accuracy of wind power forecasting.Finally,the results of case studies are given to demonstrate the effectiveness of the proposed method.

关 键 词：风电预测 岭回归 K近邻算法 梯度下降 纵向联邦学习 分布式优化 

分 类 号：TM614[电气工程—电力系统及自动化] TP181[自动化与计算机技术—控制理论与控制工程]