Wind Turbine Fault Early Warning Based On SCADA Data

Wind turbines have been operating in harsh environments such as shearing wind,sandstorms,thunderstorms,and self-shocks.Wind turbines are prone to various failures.If a serious failure occurs,the wind turbine will be forced to shut down unplanned,which will bring huge economic losses to the wind farm.It is of great academic significance and engineering application value for wind farms to grasp the health status of wind turbines in time,find out the potential faults of wind turbines effectively before the faults occur,and repair the faults in advance.The temperature parameters in the SCADA data of wind turbines are very important monitoring data,among which temperature abnormalities are an important indicator of mechanical faults.This article focuses on the high-frequency faults in the actual operation of the UP82-1500 double-fed asynchronous generator set produced by United Power in a wind farm in Hebei Province of Longyuan Power.Based on the characteristics of temperature change,the fault warning strategies of the generator front bearing and the main bearing are mainly studied.The fault warning strategies of the generator front bearing based on the Bayesian optimization XGBoost algorithm and the main shaft bearing of the wind turbine based on the AC-GAN data reconstruction are respectively studied.The main contents are as follows:(1)In view of the large number of wrong data and outliers in the SCADA data of wind turbines,three methods to remove outliers are proposed(Three methods based on interquartile range principle,pauta criterion(3σ criterion)and KNN algorithm),and the advantages and disadvantages of the three methods are compared through experiments.(2)A Bayesian optimization XGBoost algorithm for early warning strategy of wind turbine generator front bearing failure is proposed.The temperature prediction model of the front bearing of wind turbine generator was established based on Bayesian optimized XGBoost algorithm.The fault warning threshold of the front bearing of wind turbine generator was determined based on pauta criterion.The sliding window is used to monitor the running status of the front bearing of the generator in real time to improve the accuracy of the early warning.The experimental results show that the proposed method can detect the abnormal signal of the front bearing of the wind turbine generator in advance.Compared with models established by random search and grid search,the advantages of Bayesian optimization model in generalization performance and prediction accuracy are verified.(3)A fault warning strategy of wind turbine spindle bearing based on AC-GAN data reconstruction is proposed.The temperature prediction model of wind turbine spindle bearing based on Light GBM is established by using SCADA data.The normal and abnormal temperature residuals of the spindle bearing of abnormal wind turbine are distinguished by sliding window and SPC method.In order to improve the accuracy of fault sample marking,AC-GAN is used to generate the residual data which is similar to the abnormal residual data in the temperature residual distribution of the abnormal spindle bearing of wind turbine.A state decision model based on NGBoost is established to judge the state of wind turbine spindle bearings,which solved the problem of low generalization caused by using a single fixed threshold or setting a threshold subjectively in the traditional method for wind turbine operation state monitoring.The experimental results show that the proposed method can detect the abnormal signals of wind turbine spindle bearings in advance,and the average accuracy of the state decision model based on NGBoost is improved from 60.5%(Residual-free sequence reconstruction)to 72.3%(Residual sequence reconstruction).