Research On Fault Diagnosis Of Wind Turbine Based On Convolutional Neural Network And Ensemble Learning

As a renewable and clean energy,wind energy is increasing its share in the world market.However,wind farms are usually located in remote areas and wind turbines have been eroded in harsh environment for a long time.With the increase of service life,the performance of wind turbines will decline and frequent failures occur,which will not only produce more expensive maintenance cost,but also cause serious safety accidents.Therefore,the fault diagnosis of wind turbine is of great significance to reduce the operation and maintenance cost and improve the operation safety.In order to solve the problem of multi-category fault detection and identification in the process of wind turbine operation,based on the operation status and fault record data sets provided by the supervisory control and data acquisition system and the condition monitoring system of wind turbines,this paper conducts multi-category fault diagnosis research on multiple wind turbines based on feature selection research and convolutional neural network combined with ensemble learning algorithm.Firstly,this paper reviews the research background of wind turbine fault diagnosis,introduces the research literature based on model,signal and data-driven,analyzes the research objects and features of various methods.Based on the data sets of wind turbines with low sampling frequency,high feature dimensions and multiple state types,the basic research route is determined to use feature selection to optimize feature sets,then use convolution neural network to design a single wind turbine fault diagnosis model,and finally combine clustering analysis and ensemble learning to build a multi-wind turbine fault diagnosis system.To explore the correlation between data features and operation status,this paper implements feature selection on wind turbine operation status data sets.By using variance threshold,chi square-test and Pearson correlation coefficient in the filter method,combined with recursive feature elimination and random forest in the wrapper method,the feature subset which is most closely related to the operation status of wind turbines is established.Aiming at the wind turbine data sets with low sampling frequency and small fault data samples,this paper presents a fault diagnosis model of single wind turbine based on improved convolutional neural network.The model uses convolution and pooling to extract features from the original data,adjusts neurons and convolution kernel,adds hidden layers,and improves activation function and learning rate of the traditional network.The experimental results indicate that the diagnosis results are significantly improved.For multiple wind turbines in the same wind farm,the clustering analysis is carried out and the clustering data sets are generated.The fault diagnosis model proposed in this paper is used for experiments.The results show that the effect of fault diagnosis using clustering is better than that of single wind turbine.Furthermore,the ensemble learning algorithm is introduced and combined with single convolution neural network model,a multi wind turbine fault diagnosis model based on ensemble learning is proposed.The experimental results illustrate that the proposed method can significantly improve the fault diagnosis performance of multi wind turbines in the wind farm.