Research On Fault Diagnosis Of Wind Turbine Generator Based On Deep Learning

Renewable energy has attracted considerable interest globally in recent years,becoming a viable option due to technological developments,reduced costs and increased demand,especially in developing countries.Among them,wind power generation is one of the most promising renewable energy technologies that has developed rapidly in recent years,providing a significant share of electricity in an increasing number of countries.With the installed capacity of wind power increasing year by year,the wind turbine equipment is easy to break down due to the bad operating environment and complex working condition of wind turbine.Due to the failure of the generator system,which not only greatly increased the cost of operation and maintenance,but also greatly hindered the sustainable development of the wind power industry.In order to improve the reliability of wind turbine operation,timely diagnose the unit failures caused by mechanical and electrical failures,optimize the unit operation and maintenance strategy,This paper is based on the Supervisory Control and Data Acquisition(SCADA)system data of wind turbines in actual operation,with deep learning theory as the main technical theory support and wind turbine generators as the main research object.Fault diagnosis of wind turbine generator based on deep learning is proposed.The main work completed is as follows:Firstly,the SCADA data of wind turbine generator set is large,the nonlinear is strong,and it is difficult to dig deep information in ordinary network.This paper presents a Normalization network based on Group Normalization(GN)and the Stacked Denoising Auto Encoder(SDAE).This network is based on the SCADA data of the operation of wind turbine generators.In order to solve the problems of slow training and convergence of stacked noise reduction self-coding network,a group standardization algorithm is added.Then,RMSProp algorithm is used to adjust the weight and bias of the self-encoder,so as to improve the situation that the oscillation range of loss function is too large in the updating process.Finally,the softmax activation function is used in the last layer of the network to classify the results,so that the output results of the whole network into a probability distribution mode.The SCADA data of selected wind turbines were substituted into the stacked noise reduction self-coding network before and after improvement respectively for comparative training verification.The results show that the stacked noise reduction self-coding network based on group standardization is more accurate and effective for the status monitoring and fault diagnosis of wind turbines,and also provides a reference for the fault identification of wind turbines.There are few fault data in the original SCADA data of wind turbine generators,so only a small number of fault samples can be used for fault diagnosis.As a result,the training and learning of deep learning network are not sufficient,and it is difficult to accurately learn the mapping relationship between characteristic parameters and fault types from the original data.As a result,it is difficult to improve the fault diagnosis performance of deep learning network.A fault diagnosis method based on improved Auxiliary Classifier Generative Adversarial Network is proposed to realize the fault diagnosis of wind turbine generator.Firstly,the original discriminator of ACGAN network was replaced by GN-SDAE network,and the ACGAN-GN-SDAE network was built.Secondly,due to the strong ability of ACGAN network original generator to expand small sample data and the superior ability of the replacement GN-SDAE network discriminator to mine data features,the ACGAN-GN-SDAE network has a high fault diagnosis accuracy through the zero-sum game adversary-training mechanism.Finally,the improved ACGAN network was compared with the ordinary SDAE network,GN-SDAE network and other networks,and the experimental results verified the effectiveness of the improved ACGAN network in the fault diagnosis of wind turbine generators.