Research On Condition Monitoring And Fault Diagnosis Of Wind Turbine Based On Multi-source Data

As countries around the world pay more attention to controlling the total amount of carbon dioxide emissions,wind power,as a mature new energy,plays an important role in the process of "energy conservation and emission reduction".In recent years,China’s wind power industry has developed rapidly,and the installed capacity has increased year by year.However,the operating environment of wind turbines is harsh,most of them are located in remote areas and mountainous areas,and equipment failures occur frequently,seriously affecting the operating efficiency of the units and the security of the power grid.In view of the above background and current situation,this paper starts from the selection of data characteristic variables,and researches on condition monitoring under variable working conditions,fault diagnosis of Hidden Markov Model(HMM),and comprehensive utilization of various data of wind turbine generators.The main contents are as follows:(1)In order to further improve the correlation of feature variable selection results of SCADA data,a feature variable selection method based on D-S evidence theory is proposed.The filtering,wrapper and embedded feature selection methods are used to calculate the weight coefficients of each feature variable corresponding to different methods.The advantages of D-S evidence theory are given full play,and the selection results obtained by the above methods are comprehensively considered according to certain rules.According to the influence of different selection methods on model accuracy,the feature variable selection method based on D-S evidence theory can comprehensively and accurately select the feature variable highly related to the target variable from SCADA data.(2)Aiming at the problem of the accuracy of state monitoring of wind turbines under variable working conditions,a method of monitoring its operation conditions using state curves is proposed.First,the five state curves of wind speed-power,wind speed-rotation speed,wind speed-pitch angle,speed-power,and speed-pitch angle under different operating stages of wind turbines are analyzed theoretically.Then use the normal operation data to determine the distribution diagram of each state and clean it through the Mahalanobis distance,and further compare and study with the theoretical curve.By comparing and selecting the state curves of speed-power and speed-pitch angle as the monitoring basis,the deviation threshold under each working condition is determined,and the monitoring strategy is established to effectively monitor the abnormal state under different working conditions and improve the monitoring accuracy.(3)In order to solve the problem that the discrete hidden Markov model does not fully use the rich feature information of input signals in the process of fault diagnosis,a fault diagnosis method based on hidden Markov model and convolutional neural network(CNN)is proposed.The fault diagnosis strategy of discrete hidden markov model is studied.On this basis,CNN is used as the transmission probability matrix of HMM,the probability of each state to the input continuous signal is calculated,the CNN HMM model is established,and historical data is used for training.The example shows that the CNN-HMM model has better identification accuracy than the discrete HMM model,and the separation between different faults is more obvious.(4)Aiming at the problem that SCAD A data and vibration data are not fully used in the condition monitoring and fault diagnosis of wind turbine,a method of condition monitoring and fault diagnosis of wind turbine based on multi-source data is proposed.Firstly,SCADA data is used to establish the state curve for state monitoring,and then CNN-HMM model is used for fault diagnosis.During normal operation,the Gaussian mixture model is used to identify the working conditions of wind turbines,and the best model is selected for multi working conditions monitoring;After the abnormal condition is detected,the kernel Principal Component Analysis(KPCA)is used to fuse the SCADA data,the time-domain and frequency-domain characteristic variables of the vibration signal for CNN-HMM fault diagnosis.The simulation results using field operation data show that this method can not only timely monitor the abnormal state of wind turbines,but also the fused sample data can reflect the off design condition of wind turbines,with high identification of fault characteristics,and effectively improve the safety and stability of wind turbine operation.