基于神经网络和稳健估计的风电机组状态监测

打开文本图片集

中图分类号：TP277DOI：10.3969/j.issn.1004-132X.2025.08.019 开放科学（资源服务）标识码（OSID）：

Abstract： In the condition monitoring of wind turbines，temperature time-series data was used as a key indicator to evaluate the stability of their operations，typicallycolected by the supervisory control and data acquisition（SCADA） systems.A new method was proposed that leveraged temperature data for more robust wind turbine condition monitoring. To addressthe slow convergence issues in traditional prediction models，a network structure combining CNN and BiGRU was adopted，and a novel optimization algorithm —COA was introduced，to improve the training performance of the temperature prediction model. Furthermore，considering the high false alarm rate of traditional control charts in actual operational environments， a strategy was proposed that integrated median estimation（MED） and minimum regularized weighted covariance determinant（MRWCD） for robust monitoring of residual vectors.Based on these improvements，a multivariate exponentially weighted moving average control chart was established.The applications in a wind farm located in east China demonstrate that，compared with traditional monitoring methods，the pro posed approach reduces false alarms significantly and provides higher reliability and stability in wind turbine condition monitoring.

Key words：wind turbine condition monitoring；convolutional neural network-bidirectional gated recurrent unit（CNN- BiGRU）；coati optimization algorithm（COA）；robust test statistics

0 引言

能发电成为推动国家可再生能源发电的重要组成部分[2]。（剩余16375字）