Wind Turbine Fault Diagnosis And Early Warning Based On Operation Data

With the increasing of total energy demand,wind energy has become a hot spot in the field of new energy with its advantages of wide distribution,easy access and no pollution.The use of wind turbines for wind power generation is a common way to utilize wind energy,but operating under harsh environment and complex working conditions for a long time,wind turbine is very easy to receive impact causing damage to its components,which makes the maintenance process very difficult and affects the economic benefits of wind farms.Therefore,ensuring the safe and stable operation of wind turbines has become the focus of research in the wind power industry.This paper study the fault diagnosis and fault earlying warning method of wind turbines based on operational data,through in-depth analysis of the operation data of wind turbines,the types of faults that have been generated and the maintenance direction is clarified;warning potential faults to avoid serious faults of the turbines.Aiming at the problem of lack of wind turbine fault data and data imbalance affecting the fault diagnosis accuracy,the improved SMOTE algorithm of WKFCM is used to synthesize data,reduce the imbalance degree of data sets and realize the expansion of fault data sets.Based on the idea of random out-of-bag error,the importance of feature is sorted,and the most relevant features are filtered as input parameters to reduce the complexity of the model.The random forest fault diagnosis model is established,and the model parameters are optimized by using the grid search algorithm.Through the fault diagnosis confusion matrix,it can be seen that the random forest fault diagnosis model can accurately classify multiple operating states of wind turbine.In terms of fault warning,a comprehensive correlation indexes is used to select parameters highly correlated with gearbox oil temperature and output power,and then CNN network is combined with Bi-LSTM network to establish CNN-Bi-LSTM parameter prediction model to achieve in-depth mining of the spatial information and time information of the sample data.To calculate the residual value of actual value and predicted value,the residual distribution fit is carried out by kernel density estimation,abnormal warning threshold and fault warning threshold are set based on the interval estimation idea,and the sliding window is used to smooth the prediction data to eliminate accidental errors,the wind turbine is warned according to the smoothing residual.Through the actual operation data to verify the effectiveness of the fault early warning model,the results show that the residual value fluctuates within the allowable range in the normal state,and the residual value in the fault state obviously exceeds the fault warning threshold,the fault warning can be carried out 1.5 h in advance to reflect the abnormal operation of wind turbines and improve the reliability of wind turbine operation.