Fault Prediction Of Wind Turbines Based On Operating Data

In recent years,the wind power industry has developed rapidly in our country.More and more wind farms have been built,and a large number of wind turbines have been connected to the power grid for generating electricity.However,the failure rate of wind turbines is an unignorable problem,due to that faults of wind turbines may cause severe damages to the wind turbines,and the cost of maintenance is usually very high.Therefore,fault prediction of wind turbines has become a critical issue deserved to study.Based on the principles of wind turbine operation,as well as the wind data,considering the gearbox high-speed shaft bearing overheating fault as the research object,this paper employ several data mining methods to analyze the faults of wind turbines based on historical operation data from a SCADA system.Specifically,we make the following contributions:Firstly,we present a feature selection method for filtering irrelevant variables using professional knowledge,and selecting variables that can reflect the behavior of the faults of wind turbines.Meanwhile,based on ReliefF algorithm,the feature is selected,and correlation variables are selected to build the model based on correlation analysis,which lays the foundation for the validity of modeling data.Secondly,in order to determine whether the gear box high-speed shaft bearing is in overtemperature state,we employ the SOM self-organizing neural network and support vector machine(SVM)to build fault prediction models.For SOM self-organizing neural network method,the selection of no category tag sample training model,it can cluster data effectively,and then through compared with the actual category sample to achieve the purpose of sample identification.For SVM method,the SVM model is selected to train the SVM model with class labeling of both normal and fault classes,and the classification ability of SVM is used to identify whether the sample is a normal sample or a fault sample.Through modeling after data processing and the original training data set,the test set validates the model and finally uses the confusion matrix to evaluate the performance of the model.Finally,we compare and analyze the two models constructed by SOM self-organizing neural network and support vector machine(SVM).It is shown that SVM outperforms SOM significantly,and thus it can be considered an appropriate method for fault analysis and prediction for wind turbines,so as to achieve the target of improving the wind turbines operation stability and availability.