Research On Condition Monitoring Of Key Components Of Wind Turbine Based On SCADA Data

With the rapid development of wind power industry,the fault diagnosis and condition monitoring of wind turbine lead to high operation and maintenance costs.It is a difficult problem for wind power development to improve the accuracy of wind turbine condition monitoring and identify early abnormal accurately.The supervisory control and data acquisition(SCADA)system provides the status data in the operation of wind turbine.It is unnecessary to install additional sensors for the condition monitoring of wind turbine by using SCADA data.Based on the rich status information in SCADA,this paper studies the condition monitoring of three key components of wind turbine(gearbox,generator and tower).The main work includes:(1)the temperature anomaly monitoring of gearbox and generator is studied based on time window denoising auto-encoder;(2)The abnormal monitoring and trend analysis of tower vibration.Aiming at the research of gearbox and generator temperature monitoring,an abnormal monitoring method based on time window denoising auto-encoder is proposed.The time window denoising auto-encoder is introduced to fuse the state information of various sensor data in time dimension and space dimension,and the abnormal state recognition threshold is determined based on kernel density estimation method.According to the residual distribution of the time window denoising auto-encoder,the abnormal state recognition of gearbox and generator is carried out.Finally,it is verified by SCADA data.The results show that considering the influence of historical operation status and fusing multiple sensor information,the accuracy of anomaly monitoring is improved,which is conducive to early fault identification.For tower vibration monitoring research,firstly,the influence of wind turbine operating conditions on tower vibration is analyzed,and a condition identification method based on kmeans clustering is proposed.The derived characteristics are constructed according to the operating characteristics of wind turbine,and the kmeans algorithm is introduced to carry out the accurate division of wind turbine operating conditions.Then,the correlation coefficient and grey correlation analysis are used to analyze the tower vibration,and a correlation evaluation method based on denoising auto-encoder is proposed to analyze the influence of different state parameters on tower vibration.The results show that the vibration of the tower is quite different under different conditions,and there is a strong correlation between pitch and tower vibration.Finally,a tower vibration prediction model based on extreme gradient boosting(XGBoost)is proposed.Considering the influence of conditions on tower vibration,the state parameters related to tower vibration are selected to establish the model to predict the vibration trend of tower.A tower vibration monitoring and evaluation method based on Wasserstein distance is proposed.The tower state is evaluated by comparing the distribution differences of tower vibration prediction residuals of different units in the same wind farm,and the tower vibration prediction residuals in different stages are analyzed.The results show that the proposed method can accurately predict the tower vibration trend and effectively monitor the abnormal tower vibration.