摘要
针对火电厂标注样本缺失且源域无关类样本导致用传统领域迁移方法难以实现有效迁移的问题,首先建立基于双向门控循环单元的特征提取器,并利用梯度反转层引入域对抗训练机制,协同最大均值差异损失共同缩小源域与目标域之间的特征分布差异;同时,设计样本权重标定网络消除源域不相关样本造成的负迁移影响;最后,利用三元组损失增大源域与目标域数据中不同类故障之间的特征距离,显化特征决策边界。仿真实验表明,用该方法可有效实现对类别未标定目标域的特征迁移,并且在源域存在无关类样本时仍能保证对目标域较高的故障诊断精度。
Aiming at the problem that the traditional domain migration method is difficult to achieve effective migration due to the lack of labeled samples and source domain independent samples in thermal power plants,a feature extractor based on bidirectional gated circulation unit is first established,and the gradient inversion layer is used to introduce the domain adversarial training mechanism,which cooperates with the maximum mean difference discrepancy to reduce the difference of feature distribution between the source domain and the target domain;at the same time,a sample weight calibration network is designed to eliminate the negative migration effects caused by unrelated samples in the source domain;finally,the triple loss is used to increase the feature distance between different types of faults in the source domain and the target domain data,so as to visualize the feature decision boundary.The simulation experiments results show that this method can effectively realize the feature migration of the target domain with uncalibrated categories,and can still ensure high fault diagnosis accuracy for the target domain when there are unrelated class samples in the source domain.
作者
马良玉
韩立凯
MA Liangyu;HAN Likai(Department of Automation,North China Electric Power University,Baoding 071003,China;Key Laboratory of State De and Optimal Regulation for Integrated Energy System,North China Electric Power University,Baoding 071003,China)
出处
《电力科学与工程》
2025年第6期50-57,共8页
Electric Power Science and Engineering
基金
河北省中央引导地方科技发展资金资助项目(226Z2103G)。
关键词
故障诊断
对抗训练
自适应权重
最大均值差异
三元组损失
fault diagnosis
adversarial training
adaptive weighting
maximum mean discrepancy
triplet loss
