Bearing Fault Warning Of Wind Turbine Based On SCADA Data

As a kind of renewable clean energy,wind energy is the most promising renewable energy with huge reserves,wide distribution and low development and utilization cost.However,wind turbines are mostly located in areas rich in wind energy resources,such as the gobi desert,wilderness,mountain,etc.,the operating environment is relatively harsh,which will cause certain impact on the operation of wind turbines.At the same time,long-term operation will also cause certain wear and tear on the inside of the fan.Under the influence of various factors,the parts in different parts of the fan are prone to a variety of faults,among which the wind turbine bearing fault is the most common.The bearing failure of the wind turbine may cause low working efficiency of the wind turbine,may cause the wind turbine to stop,and may even lead to the complete scrapping of the wind turbine,which will seriously affect the production safety and economic benefits of the wind farm.Therefore,it is of great significance to study the bearing fault warning of wind turbine for the safe production of wind farms.SCADA(supervisory control and data acquisition)system is a supervisory control and data acquisition system installed in a wind farm.In this paper,based on the data collected by SCADA system,machine learning method is used to analyze and study the bearing over-temperature faults of wind turbine.The main tasks include the following:(1)Aiming at the problem of data preprocessing of wind turbines,this paper proposes a data preprocessing method based on improved bin algorithm,which can effectively clean the collected original data and remove abnormal data.There is a lot of noise in the original data of wind speed and power collected by SCADA system,which brings great challenge to the follow-up application research.Based on the spatial distribution characteristics of the wind speed-power data,this paper divides the wind speed-power data into three categories,and proposes a cleaning method and process for identifying abnormal data based on the sub-region bin(dbin)algorithm.Experimental results show that the dbin algorithm is more efficient than the traditional algorithm in identifying abnormal data and has a strong universality.(2)Aiming at the bearing over-temperature problem of wind turbine,this paper constructs the bearing over-temperature early warning model from two different angles and puts forward the evaluation index.The algorithm is based on the regression parameters,the use of the normal history data to construct the bearing temperature curves of normal model reference curve,using the data to construct historical over-temperature bearing curves of over-temperature,adopt the method of average distance of abnormal and normal reference curve regression curve,to quantify the degree of deviation of reflected bearing potential over-temperature problems of health indicators unit.The second is to build an early warning model of bearing over-temperature fault based on SVR(Support Vactor Regerssion)algorithm: the predicted value of bearing temperature is obtained by training the historical data of bearing temperature,power and speed.The deviation between the predicted value of bearing temperature and the actual measured value is calculated,and the threshold value is compared to determine whether the bearing is over temperature,so as to achieve the purpose of early warning of bearing over temperature of wind turbine.(3)Aiming at other bearing faults of wind turbine,this paper constructs a bearing fault identification model based on self-paced learning algorithm.In the training process of self-paced learning,the samples with high likelihood value and small training error are firstly selected for iteration,and the model parameters are constantly updated until all samples are iterated or the loss function is reduced to the minimum.The self-paced learning method is used to analyze the data of 9 faults,such as the bearing’s torsion angle deviation fault,the bearing’s torsion angle gain fault,and the bearing’s torsion angle stuck fault,etc.,and compared with LVQ(Learning Vector Quantization)algorithm,ELM(Extreme Learning Machine)algorithm and decision tree algorithm,it can be concluded that the self-paced learning algorithm has the highest classification accuracy and the best effect.Based on the data collected by SCADA system and the machine learning method,this paper studies the bearing over-temperature fault of wind turbine,the torsion angle deviation fault of wind turbine bearing,the torsion angle gain fault of bearing and the torsion angle stuck fault of bearing.The experimental results show that the method presented in this paper is effective in early warning of bearing failure of wind turbine.