Condition monitoring and anomaly detection of wind turbine based on cascaded and bidirectional deep learning networks

• Published in: Applied energy Vol. 305; p. 117925
• Main Authors: Xiang, Ling, Yang, Xin, Hu, Aijun, Su, Hao, Wang, Penghe
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2022
• Subjects: Attention mechanism; Cascaded deep learning networks; Condition monitoring; Quartile; Wind turbine; Condition monitoring; Quartile; Cascaded deep learning networks; Attention mechanism; Wind turbine
• ISSN: 0306-2619; 1872-9118
• DOI: 10.1016/j.apenergy.2021.117925

•A novel model is proposed for data-driven condition monitoring of wind turbine.•Two directional GRUs (BiGRU) are proposed to extract information from back to front.•The quartile method is developed to distribute the SCADA data for data preprocessing.•The proposed CNN-BiGRU-AM model can provided guarantee for wind energy application. Renewable energy is widely applied in the world. The key problem of wind energy application is to improve the reliability of wind turbine and reduce its downtime. Supervisory control and data acquisition (SCADA) has created reliable and cost-effective status data for health conditioning of wind turbine operation. Effectively extracting useful information from SCADA is critical to the reliability of applied wind energy. In this paper, a new method is proposed to extract multidirectional spatio-temporal features of SCADA data for wind turbine condition monitoring based on convolutional neural network (CNN) and bidirectional gated recurrent unit (BiGRU) with attention mechanism. Firstly, the quartile method is developed to distribute the SCADA data for cleaning and deleting the abnormal data so as to improve the data validity. Then, the input variables are selected through Pearson correlation coefficient, and they are transformed into high-dimensional features by using CNN. These features are input into BiGRU network through attention mechanism layer. Attention mechanism strengthens the impact of important information to improve learning accuracy. In the end, it is verified that the proposed method can detect early abnormal operation and identify failed components of wind turbine by real case analysis from wind farm. The proposed method presents better feasibility of practical wind energy application, which can promote the application of clean energy.