海上风电场集电线路故障区段定位迁移学习方法  

Transfer learning for fault section location of collector lines in offshore wind farms

作　　者：白通 高玉青 杨林刚 周才全 王慧芳[1] BAI Tong;GAO Yuqing;YANG Lingang;ZHOU Caiquan;WANG Huifang(College of Electrical Engineering,Zhejiang University,Hangzhou 310027,China;Power China Huadong Engineering Corporation Limited,Hangzhou 311122,China)

机构地区：[1]浙江大学电气工程学院,浙江杭州310027 [2]中国电建集团华东勘测设计研究院有限公司,浙江杭州311122

出　　处：《能源工程》2024年第6期100-106,共7页Energy Engineering

基　　金：宁波市“科技创新2025”重大专项项目(2018B10024);浙江大学宁波科创中心项目(2021-KYY-508101-0168)。

摘　　要：准确定位集电线路故障区段是海上风电场集电海缆故障快速恢复的关键,对于保障海上风电场的经济利益具有重要意义。在海上风电场实际运行中,可能会发生各种未知故障,集电线路的拓扑也会发生变化,基于数据驱动的集电线路故障区段定位方法在实际风电场中的应用效果会变差。针对这一问题,提出了一种基于领域对抗和图卷积神经网络的集电线路故障区段定位方法。在新的应用场景下,可缩小旧数据和新数据之间的分布差异,所提取的公共特征能够使模型对新场景具有良好的适应性。仿真结果表明,所提方法比传统机器学习方法定位精度更高,领域对抗迁移学习机制显著提高了模型在未知场景中的适应能力。Accurately locating the faulty section of collector lines is the key to the rapid recovery of collector sea cable faults in offshore wind farms,which is of great significance for safeguarding the economic interests of offshore wind farms.In the actual operation of offshore wind farms,a variety of unknown faults may occur,and the topology of the collector lines may change,and the data-driven collector line fault segment localization method based on the data will deteriorate in the application of actual wind farms.To address this problem,a collector line fault section localization method based on domain confrontation and graph convolutional neural network is proposed.In the new application scenario,the distribution difference between the old data and the new data can be reduced,and the extracted common features can make the model well adapted to the new scenario.Simulation results show that the proposed method has higher localization accuracy than traditional machine learning methods,and the domain adversarial migration learning mechanism significantly improves the model's adaptability in unknown scenarios.

关 键 词：故障区段定位 海上风电场 集电线路 迁移学习 领域对抗 

分 类 号：TM773[电气工程—电力系统及自动化]