Data-based Health Assessment And Fault Prediction Of Wind Turbine Generator

The generator is a key component of wind turbine.The fault frequency of generator is relatively high and the fault maintenance takes a long time during the actual operation.Therefore,it is of great significance to study the health management of generators and improve the intelligent maintenance of wind turbines to ensure the safe and economic operation of wind turbines.This paper mainly studies the health assessment and fault prediction of generators.The main contents are as follows:(1)The selection of parameters is important for the health assessment of generator.At present,most of the current research only relies on the expert experience,which leads to the great influence of human factors.This paper proposes a health assessment method of generator based on correlation analysis.First,the Pearson correlation coefficient and MIC are used to fully evaluate the correlation and redundancy between state parameters.The state parameters related to generator are extracted from the SCADA system and the redundant state parameters are removed.Then,a health benchmark model based on Gaussian Mixture Model(GMM)is established using the historical data in normal operation.On this basis,a health degradation index(HDI)based on Mahalanobis distance is designed to evaluate the online operating state of generator.Finally,the method is verified by the actual SCADA data.The test results show that the HDI can track the change of generator operating status accurately and play a good role in early fault identification.It can also reduce the number of leak alarm obviously.(2)Considering the complex and non-stationary operation environment,a health assessment model of generator based on operational condition recognition is proposed.First,the operational conditions are divided using the operational parameters data and the K-Means clustering algorithm.Then,after solving the data imbalance,a health benchmark model is constructed in each operational condition.An operation recognition model is trained and the final health degradation index(HDI)is constructed.Furthermore,considering the non-stationary of the online operating state,an alarm method is designed based on global fixed threshold and local dynamic threshold.Finally,the effectiveness of the method is verified by the actual data.The HDI after operational condition recognition is more sensitive to the change of operating state than the HDI without operational condition recognition.At the same time,it can reduce the number of leak alarm and fault alarm.(3)The generator rear bearing temperature alarm is a critical and frequent fault of generator.This paper proposes generator rear bearing temperature fault prediction models based on Support Vector Regression(SVR)to predict the future temperature and the remaining time of the fault.First,the moving average method is used to overcome the random variation of the online data.Considering the different influence of each state parameter on the generator rear bearing temperature fault,the weight of each state parameter is designed using the correlation analysis results.The grid search and cross-validation are used to optimize the parameters of the SVR model to improve the accuracy.The validation results show that the accuracy of the proposed method is better than the other three methods.