期刊
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
卷 18, 期 7, 页码 4698-4706出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TII.2021.3115567
关键词
Feature extraction; Correlation; Task analysis; Convolution; Convolutional codes; Power quality; Informatics; Deep learning (DL); label correlation; multilabel classification; power quality disturbances (PQD)
类别
资金
- National Natural Science Foundation of China [51777061]
In this article, a novel multilabel method named LGAN is proposed, which incorporates deep learning and explores the correlations between PQD labels to improve performance. Comparative experiments show that LGAN outperforms existing multilabel methods in terms of PQD.
Multiple power quality disturbance (PQD) contains various single disturbances, so its classification is essentially a multilabel classification task. Due to the complexity of multilabel tasks, the performance of existing multilabel methods for PQD is hard to meet practical needs, which severely restricts the engineering application of multilabel methods. Therefore, in this article, we propose a novel multilabel method named LGAN, which incorporates deep learning and explores the correlations between PQD labels to improve performance. First, 1-D convolutional neural network extracts features automatically from PQD signals. Then, a label-guided attention module is adopted to learn the specific feature representation of each PQD category. Finally, the bidirectional recurrent neural network models the label correlations from the label-related features, and then predicts the final PQD category. Various comparative experiments show that the performance of LGAN is much better than the existing multilabel methods. Additionally, the test using real-time detection platform further verifies the availability of the proposed method.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据