Fault Prediction Of Wind Turbine Gearbox Based On Variable Weight Combination Model

With the increasing demand of the global energy and serious destruction of carbon emissions to the natural environment,wind energy has become an important part of renewable energy such as water,wind,solar and tidal energy.The scale of wind power industry in China has expanded rapidly,which makes the maintenance of wind turbine need to be solved urgently.Once wind turbine breaks down,it will not be diagnosed and repaired in time because of the restriction of installation environment.Besides,serious failures of wind turbine will lead to significant economic losses due to downtime maintenance.Gearbox as an important part of wind turbine has high failure rate and causes long downtime of wind turbine.Therefore,it has practical significance for fault prediction of gearbox so that wind farm can implement measures to prevent further deterioration of the gearbox.The work of this paper is as follows:(1)The structure of wind turbine is analyzed,especially the gearbox.According to working principle of gearbox,this paper puts forward the oil temperature and shaft temperature which are closely related to the normal working state of gear box.It can be used to predict fault of gearbox.The variables which are related with oil temperature and shaft temperature such as wind speed,power,inlet oil temperature are obtained in the data acquisition and monitoring system of the wind farm.(2)Constructing gearbox temperature model predicts oil temperature and shaft temperature for gearbox fault prediction.Firstly,according to the relationship of wind speed and power,the grid method and the K-mean clustering algorithm are used to eliminate the abnormal data to obtain indirectly the training data about the normal state of the gearbox.Secondly,considering the inertia of the temperature the time series method is used to analyze the internal mathematical structure of the oil temperature and shaft temperature of the gear box.Thirdly,the related variables with oil temperature and shaft temperature of the gear box are too much to analyze easily.Considering the sparsity of the high-dimensional data space,the principal component analysis is applied to reduce the dimension of the data.Finally,the improved BP neural network and grey prediction model are adopted to build the gearbox temperature model.In order to avoid the backward feedback neural network falling into local extreme value and slow speed of convergence,bat algorithm isapplied to improve it.(3)Compared with the limitation of the single prediction algorithm,the fixed weight combination algorithm and the variable weight combination algorithm are proposed to build the gearbox temperature model.Using the single prediction algorithm including improved BP neural network,gray prediction model,and ARMA model make up a combined prediction model.The single prediction model value of weight in the combination prediction model is calculated by entropy method.Considering the change of the performance of the single prediction model in the prediction of oil temperature and shaft temperature of the gearbox,the variable weight combination forecasting model is proposed based on the grey correlation degree to built temperature model of the gearbox.Comparing different single prediction model and combined prediction model,the variable weight combination prediction model is selected to predict the oil temperature and the shaft temperature of the gearbox.According to the prediction error,this paper use large sliding window to analyze and predict the gearbox fault.