Fault Diagnosis Of Wind Turbine Based On Generative Adversarial Network

Under the dual pressure of energy crisis and environmental protection,developing renewable energy has become the trend of sustainable development.Wind power generation plays an important role in new energy and it has been developing rapidly in recent years.Wind turbine works in harsh environment such as sandstorm and blizzard for a long time.Its components are vulnerable to aging and wear under the influence of complex environment,which will directly affect the power generation efficiency of wind turbine and cause huge economic losses.Therefore,timely and accurate condition monitoring and fault diagnosis of wind turbines are of great practical significance to improve the operational efficiency of wind farms and reduce the costs of operation and maintenance.The Supervisory Control And Data Acquisition(SCADA)system of wind farm provides a large number of operational data,but most of them are normal operational data,lack of abnormal fault data.The lack of fault data leads to unbalanced sample categories,which will affect the accuracy of diagnosis model.It is of great importance to accurately diagnose faults of wind turbine in the case of missing data.In this paper,Generative Adversarial Network(GAN)is introduced into the fault diagnosis of wind turbines.GAN is employed to expand the original fault data and diagnosis methods are implemented on the basis.The main work of this paper is as follows:(1)Aiming at the problem of unbalanced fault samples,an adversarial oversampling method based on GAN is proposed.Through the adversarial learning between generator and discriminator,the generator can mine the potential characteristics of different types of fault data,and generate composite fault samples similar to the real sample distribution,so as to balance the class distribution of fault data in the sample set.The diagnosis model based one-dimensional convolutional neural network is established by using the balanced enhanced sample set.The experimental results show that the proposed method can effectively balance the fault sample set and improve the diagnosis accuracy of the diagnosis model in the case of insufficient fault samples.(2)In order to further improve the accuracy of fault diagnosis of wind turbine,a fault diagnosis method based on stacked sparse autoencoder(SSAE)and extremely randomized trees was proposed.In this method,SSAE is used to mine the feature distribution law of wind power data.Combined with the reconstruction error of each layer,fault sensitive features are extracted from high-dimensional data to construct extremely randomized trees,which can realize the accurate fault diagnosis of wind turbine.The fault data of yaw system collected by SCADA system in the wind farm are used to carry out experiments,and the results show that the proposed method has high diagnostic accuracy.