Research On Fault Prediction Method Of Switch Equipment Based On Deep Learning

In recent years,under the strategic background of a powerful country in transportation in the new era,domestic policies related to rail transit have been promulgated one after another.The railway network is constantly being encrypted and the speed of trains is also increasing,in order to ensure the safe operation of trains,railway signal equipment is under great pressure of operation and maintenance.At present,the signal equipment on the railway site is mainly based on fault repair and periodic repair,which requires the staff to have rich maintenance experience.This method has problems such as high operation and maintenance costs,low efficiency,easy to misjudgment,missed judgment,and it is difficult to adapt to the rapid development of national rail transit and the requirements for state repair and intelligent analysis of signal equipment.In response to the above situation,this thesis combines the deep learning method with the fault prediction problem of switch equipment.The main research contents are as follows:(1)The research status of fault diagnosis and prediction technology of switch equipment at home and abroad is analyzed,taking the most classic ZD6 switch equipment as the research object,the structural composition of switch equipment and the action circuit of switch are introduced.According to the study of theory and the field investigation of electrical service,seven common switch fault modes are summarized and their action current curves,fault causes and countermeasures are described.(2)The time domain features of switch action current curve are extracted by time-fixed subsection method,and the synthetic minority oversampling technique is adopted to keep the normal data and fault data of switch action current curve relatively balanced.Then the dimension of feature is reduced by Fisher criterion and kernel principal component analysis method,and finally the feature factor of slowly changing fault after dimension reduction is obtained,which lays the foundation for the training of the fault prediction model of the subsequent switch equipment.(3)The current curve during continuous operation of the switch equipment prior to failure is processed according to the method of processing the characteristic data above,and the slowly changing fault characteristic factor sequence data set is obtained,which is used as the input of the subsequent model.A switch equipment fault prediction model based on long short term memory network is proposed,for single-layer and multi-layer LSTM networks,the fixed variable method and particle swarm optimization algorithm are used to simulate experiments to obtain the best network parameters,the experimental results show that the two-layer LSTM model has better fault prediction effect.Then,gated recurrent unit neural network is proposed on the basis of LSTM and trained,the results show that the two-layer GRU model has better performance compared with the LSTM model by introducing multiple performance indicators for evaluation.The comparison with the method based on dynamic time warping distance further verifies the superiority of the deep learning method proposed in this thesis in the problem of switch equipment fault prediction.(4)Combined with the above research,using Py Charm development environment,Python programming,Py Qt5 and My SQL technology,the design of the switch equipment fault prediction system software is realized.The test data is used for verification,which proves that the system can effectively realize the function of predicting the failure of switch equipment.It can be considered as a sub-module of the next generation centralized signal monitoring system,and it also provides a new idea for the development of related products.There are 67 figures,18 tables and 66 references.