基于CNN-LSTM的风电机组异常状态检测

Abnormal state detection of wind turbines based on CNN-LSTM

风电机组的环境恶劣和工况多变导致风电机组故障频发,为了保障风电机组的可靠运行,基于数据的机组异常状态检测尤为重要。该研究提出一种基于级联深度学习预测模型的风电机组状态检测方法,首先对风电场数据采集与监视控制(supervisory control and data acquisition,SCADA)系统的数据进行预处理,并通过距离相关系数(distance correlation coefficient,DCC)分析选取输入参数;然后结合卷积神经网络(convolution neural network,CNN)和长短期神经网络(long short-term memory,LSTM)建立观测参数与目标参数之间的逻辑关系,通过均方根误差(root mean square error,RMSE)和样本熵(sample entropy,SE)对齿轮箱轴承温度预测残差进行分析,监测齿轮箱轴承温度异常变化;最后以华北某风场的SCADA数据进行算例验证,结果表明该方法能够准确检测到齿轮箱轴承温度异常,提前发现风电机组的早期故障,为风电机组安全可靠运行提供重要价值。

The bad operating environment of wind turbines leads to frequent gearbox failures.Therefore,it is particularly important to improve the reliability of wind turbines.In view of this,a wind turbine abnormal state detection method was proposed based on a deep learning prediction model and sample entropy(SE).Firstly,the wind field data acquired by the supervisory control and data acquisition(SCADA)system were preprocessed,and input parameters were selected by virtue of a distance correlation coefficient(DCC)analysis.Then,the convolution neural network(CNN)and the long and short term neural network(LSTM)were combinedly used to establish the logical relationship between the observation parameters and the target parameters.By making use of the root mean square error(RMSE)and sample entropy,the residual temperature of the gear box bearing was predicted and analysed to monitor the abnormal temperature change of the gear box bearing.Finally,the SCADA data of a wind field in north China was used for example verification.The results show that the method can accurately detect the temperature anomaly of the gearbox bearing and find the early faults of the wind turbine in advance,providing reference information to farm staffs for the maintenance of the wind turbine.