Research On Condition Evaluation And Prediction Of Wind Turbines Based On SCADA Data

With the development of science and technology,the wind power industry has been developed rapidly,and it becomes one of the main methods of power generation at present.In order to improve the economic benefits of wind power,and have put forward higher demand for safe and reliable operation of wind power generation,the abnormal state monitoring and fault prediction of wind turbine have become a hotspot of current research.In recent years,wind farms are equipped with complete supervisory control and data acquisition(SCADA)system.The SCADA system records a lot of operational state's data of the wind turbine during its working life.These SCADA data contain information about normal operation,shutdown,abnormity,fault and so on.The effective use of SCADA data to monitor and predict the state of wind turbine has important practical significance for the normal operation and maintenance of wind farm.The paper research on the abnormal state monitoring and fault prediction of wind turbine base on the SCADA data,and the main contents are as follows.(1)The quantitative method for the correlation between the state parameters of wind turbine based on the SCADA data.First,the SCADA data of wind turbine is preprocessed,and eliminate the 'dirty data'.Then,determine the range of wind speed through the correlation of wind speed,electricity and active power,and determine the SCADA data of wind turbine.Finally,the index of average correlation is put forward base on the correlation coefficient of Pearson,Spearman and Kendall,and through the index of average correlation to quantify the correlation between the state parameters of wind turbine.(2)The method about select the state parameters of wind turbine base on the back propagation neural network(Back-propagation Neural Network,BPNN).First,determine the structure of BPNN through design input layer,output layer and hidden layer of BPNN.Then the average relevance index is used to rank the influence of the state parameters on the target parameters in descending order.Finally,according to the order about index of average correlation and select state parameters are accumulated as input of BPNN model.And the input parameter set that can represent the change of target parameters is effectively selected through the accuracy of BPNN model.(3)The method about abnormal analysis of state parameters of wind turbine base on non negative weight combination forecasting is studied.Firstly,establish three kinds of single forecasting model,they are wavelet neural network(WNN),radial basis function neural network(RBFNN)and least squares support vector machine(LS-SVM).Then base on WNN,RBFNN and LS-SVM model,a combined forecasting model based on the sum of squares of prediction errors and the minimum is established.The accuracy of each single prediction model and the combined prediction model are analyzed by qualitatively and quantitatively.Finally,a criterion for determining the state parameters of wind turbines is proposed,which base on the root mean square error(RMSE)and the approximate entropy of the predict residual error.And apply the method to the analysis the abnormal state parameters in real-time monitoring of the SCADA system of the 17 wind turbine in a North China wind farm.And obtain accurate prediction results of abnormal state parameters.