Research On Fault Diagnosis And Fault Early Warning Of Wind Turbine Based On Operation Data Analysis

In recent years,due to the emergence of environmental pollution and energy shortage,countries around the world have increased the research on wind power generation technology,so the wind power industry has been rapid development.With the increasing number of old wind turbines,the failure frequency of wind turbine is also increasing.The frequent occurrence of wind turbine failures will seriously reduce the generation efficiency of wind turbine and cause huge economic losses to the wind farm.The existing maintenance methods of wind turbines are mainly divided into post maintenance and regular maintenance.Post maintenance is more passive,which requires a lot of time and manpower to locate the fault components and fault types.The way of regular maintenance is relatively rigid,which is prone to the inconsistency between the failure window period and the maintenance window period,and wastes human and material resources.Therefore,how to make good use of the existing supervisory control and data acquisition(SCADA)system to carry out accurate fault diagnosis and timely fault warning for the wind turbine is of great practical significance to improve the maintenance efficiency of the wind turbine and reduce the failure loss of the wind farm.In order to accurately diagnose the fault of wind turbine,this thesis designs a fault diagnosis method based on ReliefF algorithm and eXtreme Gradient Boosting(XGBoost)algorithm.First of all,the method takes the SCADA system records under different working conditions of the wind turbine as the input,calculates the feature weight of each observation feature by using the ReliefF algorithm,and selects the observation feature with high degree of correlation with the fault classification of the wind turbine.Then,the selected features are taken as the features of the classification model,and the different working states of the wind turbine are taken as the labels of the classification model.The XGBoost multi classification fault diagnosis model is established to complete the mapping from the observation feature space to the different working states of the wind turbines.Finally,the real-time monitoring data of the wind turbine is input into the fault diagnosis model,and the model will automatically identify the operation status of the wind turbine.In order to verify the effectiveness of the algorithm,experiments are carried out with the actual wind farm data.The experimental results show that the recognition rate of the algorithm for seven different working states(six fault states and normal states)of the wind turbine is as high as 100%.In order to verify the superiority of the algorithm,the algorithm described in this thesis is compared with the radial basis function-Support Vector Machine(rbf-SVM)classification algorithm and Adaptive Boosting(AdaBoost)classification algorithm.The results show that the fault diagnosis accuracy of the algorithm described in this paper is higher than other algorithmsIn order to predict the failure of wind turbine in advance,this thesis proposes a fault warning algorithm based on the temperature prediction of wind turbine components.Firstly,Pearson correlation coefficient is used to select the temperature related characteristics of the wind turbine components.Then,these characteristics and the temperature value of the components in the first three moments are used as the input vector of the temperature prediction model to establish the XGBoost temperature regression prediction model of the normal wind turbine.The degree of deviation from the normal state of the wind turbine is evaluated by the residual between the predicted value and the actual value of the component temperature,and the fault warning is carried out according to the set residual threshold value.In order to verify the effectiveness of the algorithm,an experiment is designed based on the actual wind turbine data.The experimental result shows that the algorithm can detect the generator fault of the wind turbine 7.25 hours in advance and the gearbox fault of the wind turbine 2.5 hours in advance.In order to verify the superiority of the algorithm,the temperature regression prediction model designed in this thesis is compared with the Support Vector Regression prediction model.The results show that the prediction accuracy of the algorithm described in this thesis is higher.