Research On Early Warning Of Wind Turbine Gearbox Fault Based On Combined Model

In China,wind energy is mainly concentrated in the Midwest or mountainous areas with relatively poor terrain.Due to the high cost of transportation and maintenance,it is difficult to install wind turbines.Gearbox is the core component of wind turbine,and the maintenance cost and transportation cost are high,so the predictability of gearbox potential failure has become an urgent need of wind farm.To solve the above problems,this paper proposes a wind turbine gearbox fault early warning method based on combined model.Firstly,the combination forecasting model is studied.Based on the B-G combination forecasting method,according to the principle of determining the weight of combination model,five methods of calculating the weight of combination model are determined.By analyzing the limitations of the traditional combination forecasting model,this paper puts forward the improvement method of the traditional combination forecasting model,which lays a theoretical foundation for the establishment of the combination forecasting model.Secondly,two kinds of wind turbine gearbox fault early warning models are established,which are the fault early warning model based on nonlinear state estimation(nset)and the fault early warning model based on support vector machine(SVM).Markov distance is used to optimize the memory matrix of nset model,and RBF kernel function is selected to replace the high-dimensional matrix operation in SVM.After preprocessing and normalizing the sample data,the data is divided into three parts: training set,verification set and test set.The training set is used to train the model,the verification set is used to calculate the mean and standard deviation of residual error,and the threshold of model early warning is determined according to the Harcourt control chart.The test set is used to judge the fault early warning.The experimental results show that the two methods can complete the gearbox fault early warning in different degrees.Finally,the first mock exam method of wind turbine gearbox fault is proposed based on the combined model.The combined residual model is combined with the two models of the first model respectively by means of arithmetic average and optimal weight.The first mock exam is to solve the problem that traditional combination models are too dependent on single model prediction and time series.IOWA operator is introduced to optimize the combination model,so that the combined weights are only related to the prediction accuracy at a certain time.The first mock exam of various combination models shows that the IOWA combined model has the best warning effect,the optimal weight combination model is the second,and the arithmetic average combination model is poor,but it is also better than the single model.