上海交通大学学报 ›› 2025, Vol. 59 ›› Issue (5): 561-568.doi: 10.16183/j.cnki.jsjtu.2023.375

• 新型电力系统与综合能源 •    下一篇

基于零样本学习的风力机故障诊断方法

潘美琪a, 贺兴b()   

  1. a.上海交通大学 国家电投智慧能源创新学院,上海 200240
    b.上海交通大学 电力传输与功率变换控制教育部重点实验室,上海 200240
  • 收稿日期:2023-08-07 修回日期:2023-11-29 接受日期:2023-12-04 出版日期:2025-05-28 发布日期:2025-06-05
  • 通讯作者: 贺 兴,副研究员,博士生导师;E-mail:hexing_hx@126.com.
  • 作者简介:潘美琪(1998—),硕士生,从事数字孪生和故障诊断研究.
  • 基金资助:
    国家自然科学基金(52277111);上海市科学技术委员会(21DZ1208300)

A Fault Diagnosis Method for Wind Turbines Based on Zero-Shot Learning

PAN Meiqia, HE Xingb()   

  1. a. College of Smart Energy, Shanghai Jiao Tong University, Shanghai 200240, China
    b. Key Laboratory of Control of Power Transmission and Conversion of the Ministry of Education, Shanghai Jiao Tong University, Shanghai 200240, China
  • Received:2023-08-07 Revised:2023-11-29 Accepted:2023-12-04 Online:2025-05-28 Published:2025-06-05

摘要:

在工程实践中,风力机故障诊断面临训练故障与实际故障类别不同的情况,为实现对风力机未知故障的诊断,需要将训练过程中习得的故障特征信息迁移至未知故障中.不同于直接建立故障样本与故障类别间映射关系的传统方法,提出一种基于零样本学习的风力机故障诊断方法来完成故障特征迁移.通过描述每种故障的属性建立故障属性矩阵,将其嵌入故障样本空间与故障类别空间之中;并基于卷积神经网络建立故障属性学习器,基于欧氏距离建立故障分类器,形成从故障样本预测故障属性进而分类故障的诊断流程.最后通过与其他零样本学习方法的对比验证了所提故障诊断方法的有效性和优越性.

关键词: 风力机故障诊断, 零样本学习, 卷积神经网络, 知识-数据混合驱动

Abstract:

In engineering practice, wind turbine fault diagnosis encounters situations where the fault category in the training data is different from the actual one. To diagnose unknown wind turbine faults, it is necessary to transfer the fault feature information learned during training to the unknown fault category. Unlike traditional methods that directly establish mapping between fault samples and fault categories, a zero-shot learning (ZSL) method for wind turbine fault diagnosis based on fault attributes is proposed to enable fault feature migration. A fault attribute matrix is established by describing the attributes of each fault, which is embedded into the fault sample space and fault category space. Then, a fault attribute learner is developed based on convolutional neural network (CNN), and a fault classifier is established based on Euclidean distance, forming the diagnosis process where fault attributes are predicted from fault samples and then classified. Finally, the effectiveness and superiority of the proposed fault diagnosis method are validated by comparing it with other zero-shot learning methods.

Key words: wind turbine fault diagnosis, zero-shot learning (ZSL), convolutional neural network (CNN), knowledge-data hybrid driven

中图分类号: