Intelligent Fault Diagnosis And Management For Wind Turbines Based On Deep Learning

As an important part of the national energy strategy,the wind turbine is an important indicator to measure the effectiveness of the national energy strategy and improve the level of renewable energy utilization.When a large number of wind turbines are available in the market,it will be an explosive period for the operation and maintenance management services of wind turbines.According to statistics,the maintenance cost of wind turbines in 2021 is expected to exceed 30 billion yuan.With the continuous development of wind turbines in the direction of complexity,information,and intelligence,a large amount of real-time status data will be generated during the operation of the wind turbine.These massive state detection data contain valuable information that can be used for wind turbine intelligent fault management.Currently,the fault diagnosis method based on deep learning has gradually become an important technical means of current wind turbine intelligent fault management.Considering the challenges of fault diagnosis in the intelligent fault management of wind turbines and the shortcomings of existing research,this research successively expands the research of intelligent fault management methods based on non-labeled data,unbalanced data,and few-sample data.The main structure of this thesis is as follows:First,for the problem of intelligent fault recognition based on non-labeled data,a new contour density scanning clustering algorithm was proposed.This clustering algorithm achieves fast clustering without specifying the number of clusters for arbitrary shape data and is effective to the noise data.The effectiveness and practicability of the algorithm were verified by artificial synthesis data and wind turbine gearbox signal data.A case study was carried out with the bearing fault data of industrial wind turbines.The results show that the algorithm can accurately detect the faults that occur during the bearing operation and the changes of the faults over time.It provides a fault mode threshold table that can be used to monitor the running state of the bearing.A diagram of failure mode development is also provided.Secondly,for the problem of intelligent fault prediction based on unbalanced data,an improved generative adversarial neural network model was proposed.The model was designed with a fault classification prediction function retaining a data enhancement function.The optimization process fully considers the influence of synthetic data on the fault prediction results.A case study was carried out with the data of industrial blade icing faults,and the results show that the model achieved satisfactory classification prediction results.In terms of model optimization,the Nadam algorithm can better optimize the model than other algorithms,including Adam,RMSprop,SGD,and Adagrad algorithms.Then,for the problem of intelligent fault prediction with few sample data,a capsule network model based on autoencoder was proposed.This model introduces the "capsule" into the structure of the traditional autoencoder,which can effectively extract the plane features and spatial features from the few-sample data.Model optimization is achieved by reducing the losses caused by the difference between the output data and the input data and the difference between the failure prediction result and the actual failure label.A case study was carried out with the data of bearing faults under multiple operating conditions,and the results show that the model can effectively predict different failure modes under multiple operating conditions.Compared with other fewshot learning methods,this model performs better in terms of computational cost and computational accuracy.Finally,based on the above research results,the intelligent fault management process of wind turbines was introduced from two aspects of the fault management module and function implementation path,which was verified by the case of blade icing faults of wind turbines.Above all,three novel intelligent fault diagnosis models were proposed to focus on the unlabeled data,imbalanced data,and few samples in the operation of wind turbines.The intelligent fault diagnosis management modes of wind turbines were designed and the path to realizing these modes was clarified.This research expands the application of deep learning in the field of wind turbine intelligent fault diagnosis and has important guiding significance for improving the wind turbine operating environment,improving wind turbine operating efficiency,and enhancing the comprehensive competitiveness of wind power enterprises.