Research On Wind Turbine Fault Early Warning Based On Deep Auto-encoding Network

Because most wind turbines are built in windy and mountainous environments,the device failures of wind turbines are increasing.The safety issues and economic benefits of wind turbines have attracted widespread attention.Therefore,early failure warning for wind turbine is beneficial to the wind industry.Compared with classic machine learning algorithms,deep neural networks have performed better in the field of data mining.Therefore,the internal features of SCADA data can be captured by deep auto-encoder network.The fault early warning method taking a deep auto-encoder network as the core is researched.Firstly,the method based on statistical analysis and the interquartile range(IQR)-based method are used to eliminate bad data.T he processed data are normalized.Reasonable and effective data is provided for the fault warning method.Secondly,in view of the complex types and the large amount of data,the method based on the deep auto-encoder network is researched.The input parameters are selected based on the Pearson correlation coefficient method and theoretical analysis;the deep auto-encoder network early warning model is established based on offline SCADA data;the mean and standard deviation of the reconstruction error is statistically analyzed based on the sliding window model,realizing the condition monitoring of the wind turbine.The simulation result shows that the method can realize the fault warning of wind turbines.Finally,the variable conditions of wind turbines are researched,a nd the parameters of operational condition identification are determined.In view of the complex operational conditions of wind turbines,the method based on the combination of working condition identification and deep auto-encoder network is researched.The method based on the Gaussian mixture model is used to identify the operational conditions of the wind turbine.In each operational condition,the method based on the deep auto-encoding network is used to establish the warning model under normal operational condition,the multi-operating condition threshold is obtained.Analyzing the reconstruction error of the model,t he health index is constructed based on multi-operating condition threshold to realize the fault early warning.Taking the failure of a wind turbine gearbox as an example,compared with the fault warning method based DAE,this method can warn in advance.