Research On Generator Fault Diagnosis Technology Based On Monitoring Data

Wind power,as a clean and renewable energy with abundant reserves and abundant natural resources,has begun to develop rapidly under the impetus of many factors,such as the increasing lack of fossil fuel energy,energy supply safety and environmental protection.However,with the rapid development of wind power generation technology,the normal operation management and maintenance of wind power generation units,as well as fault monitoring and diagnosis and other aspects of the technology has not been rapidly improved and improved,resulting in a high probability of wind power generation unit failure.And the generator,as the core power generation equipment of the wind turbine,plays a vital role in the safe and stable normal operation of the wind turbine.If the generator failure can be accurately and effectively analyzed and diagnosed,the economic loss caused by the sudden shutdown of the wind turbine can be greatly reduced,and its normal and reliable continuous operation can be ensured.Therefore,this article mainly studies the wind turbine fault diagnosis model based on wind turbine monitoring data.Firstly,an improved stacked autoencoder algorithm is proposed.In order to analyze whether the improved stacked autoencoder algorithm can effectively diagnose the faults of generator bearings,the differences between the improved stacked autoencoder algorithm and the traditional stacked autoencoder algorithm and the advantages of using the improved stacked autoencoder algorithm are studied and analyzed through experimental comparison.The sliding KL divergence algorithm is used to diagnose the generator state reconstruction data based on the state reconstruction model of the improved stacked autoencoder algorithm.Through the design of the sliding KL divergence algorithm with appropriate window width,it can not only reflect the continuous and nonlinear changes of statistical characteristics in time and quickly,but also help to eliminate the influence of various random factors and reduce the probability of false alarm.At the same time,the reliability of the sliding window-KL divergence and the improved stacked autoencoder algorithm is demonstrated by the comparison with different methods.Then,in order to improve the accuracy of generator bearing fault classification,deep learning algorithm is used to replace the traditional feature extraction method,and a long-term generator fault classification model based on variable learning rate auto-encoding is established.For the feature extraction part,the remarkable advantages of stacked autoencoder in processing high-dimensional data and learning the internal features of complex signals are applied,and the feature model based on variable learning rate stacked autoencoder is used to extract the deep characteristic signals of generator data,which has stronger robustness and better network generalization performance.In the aspect of fault classification,combined with the excellent processing ability of the long short term memory network model for time series,the deep features of the generator bearing data extracted from the feature model were input into the long short term memory network classification model,which effectively avoided the problems such as gradient disappearance and explosion.By comparing with different hidden layers and different feature extraction methods,it is shown that the proposed method can realize bearing fault identification with high accuracy and precision.Finally,in order to alleviate the problem of poor generalization ability of single classification model,a generator fault diagnosis method based on improved ensemble learning is established,which makes the generator fault classification effect further improved.In order to prove the superiority of this method,by comparing with traditional intelligent diagnosis methods such as SVM,BP neural network and RNN neural network,it is concluded that the generator fault diagnosis method based on improved ensemble learning can achieve higher fault classification accuracy.