Fault Prediction Of Wind Turbine Based On Multiple Statistical Models

The main bearing is one of the most frequently occurring components in the direct drive wind turbine,which directly affects the working efficiency and reliability of the wind turbine.In this paper,the early failure prediction of the main bearing is carried out by using the Adaboost-BP neural network and the functional-coefficient autoregressive model combined with the exponentially weighted moving average(EWMA)control chart.The main research contents and conclusions are as follows:This paper uses the combination of Adaboost-BP neural network and exponentially weighted moving average(EWMA)control chart to predict the main bearing fault.To begin with,the monitoring data of the Supervisory Control And Data Acquisition(SCADA)system when the main bearing is working normally is selected to set up the sample training set.Based on the power curve of the wind turbine,the quartile combined with optimal intra-group variance(Quartile-OIV)algorithm is used to eliminate the derating data and downtime data in training data set.Then the main bearing temperature prediction model,which is under normal operating conditions of the wind turbine is established by using the Adaboost modified BP neural network,and the prediction residual is obtained.The residual-based EWMA control chart is established to evaluate the running state of the main bearing.This paper uses the combination of functional-coefficient autoregressive(FAR)model and EWMA control chart to predict the main bearing fault.This method takes into account the influence of the temperature of the past time on the current temperature,and it uses a variable coefficient regression model to simulate the dynamic change of temperature.Firstly,using the sequence diagram and ADF test to check the stability of the temperature data,it is found that the original temperature sequence is unstable,and the temperature series tends to be stationary after the first-order difference.Then the FAR model of the main bearing of the Wind Turbine during normal operation is establish.The EWMA control chart is used to monitor the predicted residual and to analyze the working state of the main bearing of the Wind Turbine.Finally,the data information is extracted from the operation record and fault log of a SCADA system in a wind farm in Hunan,and the extracted data is pre-processed,and then the wind turbine fault prediction model is established.The results show that the prediction accuracy of FAR model is higher than that of Adaboost-BP neural network,and the functional-coefficient autoregressive model can capture the dynamic relationship of temperature more accurately.By using the combination of two temperature prediction models and EWMA control chart,the abnormal state of the main bearing can be found and its early faults can be predicted.