Affiliations:
Hebei University of Technology, College of Artificial Intelligence, Tianjin, China
Correspondence:
[*]
Corresponding author. Dr. Liang Tao, Ph, Institute of Artificial Intelligence and Data Science, Hebei University of Technology, Guangrong Road, Hongqiao District, Tianjin, China. Fax: +8613512889536; E-mail: 54008214@qq.com.
Abstract: With the construction of large-scale wind turbines, how to reduce the operation and maintenance costs has become an urgent problem to be solved. In this paper, by extracting the actual operation data of the wind turbine in Supervisory Control and Data Acquisition (SCADA) system, the Bidirectional Recurrent Neural Networks (BRNN) is used to establish the wind turbine operation prediction model. By eliminating abnormal data points caused by accidental factors through box diagram, the fault risk threshold of wind turbine components is optimized. Then, based on the residual between the actual value and the measured value of the large sliding window, the early fault warning is realized according to Wright criterion. Finally, the model proposed in this paper is applied to the actual wind turbine, which proves the reliability and accuracy of the method.
Keywords: BRNN, box-plot, large sliding window, Letts’criterion, early fault warning
