期刊
RENEWABLE ENERGY
卷 172, 期 -, 页码 829-840出版社
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.renene.2021.03.078
关键词
Wind turbine; Anomaly detection; Long short-term memory; Auto-encoder; Mutual information theory
资金
- National Natural Science Foundation of China [51875345, 51475290]
Continuous monitoring of wind turbine health conditions using anomaly detection methods based on LSTM and AE neural network can improve reliability and reduce maintenance costs.
Continuous monitoring of wind turbine health conditions using anomaly detection methods can improve the reliability and reduce maintenance costs during operation of wind turbine. Anomaly detection aims at identifying the root causes leading to unexpected changes of product performance. Most existing methods make less use of temporal order of the data and are poor at extracting features from these data. To address these problems, a method based on long short-term memory (LSTM) and auto-encoder (AE) neural network is introduced to assess sequential condition monitoring data of the wind turbine. First, a performance assessment model is constructed using LSTM neural units and AE networks to calculate the performance indices for evaluation of the degree of anomalies in wind turbine performance. Then, an adaptive threshold estimation method based on support vector regression model is developed to identify the abnormal data instances. The mutual information theory is subsequently explored to analyze the relationships between various monitoring parameters and performance abnormal instances to identify critical condition monitoring parameters. The effectiveness of the proposed method has been verified by a case study using real-world wind turbine condition monitoring (CM) data. (c) 2021 Elsevier Ltd. All rights reserved.
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