基于降噪自编码的风力机参数异常预警研究

Research on Warning of Abnormal Parameters in Wind Turbine Based on Denoising Auto-Encoders

提出了一种基于降噪自编码的风力机参数预警方法。依据风电机组的工作原理,将其划分为风速功率系统、机舱温度系统、变桨系统、齿轮箱系统、发电机系统等子系统,基于降噪自编码(DAE)深度神经网络,建立了风电机组各子系统的参数异常预警模型。在此基础上,依据数据采集与监视控制(SCADA)系统实际运行数据质量,从空值、异常值、非正常状态数据以及归一化等4个方面提出了一套数据预处理流程,获得了训练样本,构建了风电机组降噪自编码参数异常预警方法。多台风电机组实际运行数据验证了该方法的有效性。研究成果可为风电机组的智能运行提供参考。

A warning method for abnormal parameters in wind turbine based on denoising auto-encoders(DAE)is proposed.According to the working principle of wind turbines,the whole system is divided into wind speed power system,cabin temperature system,pitch system,gearbox system,generator system,etc.Based on DAE deep neural network,a warning model for parameter abnormalities of each subsystem in the wind turbine is established.According to the actual operating data quality of the supervisory control and data acquisition(SCADA)system,a set of data preprocessing procedures are proposed from aspects of null values,abnormal values,abnormal state data and normalization.Training samples are obtained,and a warning method for abnormal parameters in wind turbines is proposed.The effectiveness of the method is verified by actual operation data of many sets of wind turbines.The research results can provide reference for the intelligence operation of wind turbines.