| In recent years,the speed and density of rail transportation have increased,and the reliability of railway infrastructure is facing serious challenges.As one of the important railway electrical infrastructures,the operation of switch machine directly affects railway service,safety and maintenance costs.Therefore,the first task of switch machine maintenance is to carry out intelligent monitoring and early warning,so as to reduce operation and maintenance costs while maintaining the safety of switch machine equipment.However,the fault data of various railway equipment is not easy to obtain,and the research of machine intelligence algorithm for fault diagnosis and prediction of switch machine lacks sufficient data support.In order to solve the above problems,this paper first designs a switch machine simulation data generator,based on Smote,Borderline-Smote,kriging methods simulation data generator can generate switch machine’s normal power data,fault data and fault progress data;Secondly,in order to construct the prediction model further,this paper takes the retardation fault as the research object of fault prediction,preprocesses the fault progress data and constructs the degradation performance index through the KPCA method.Finally,based on the LSTM neural network,the improved model PSO-LSTM is constructed,and the LSTM model before and after the improvement is compared by using the prediction model evaluation index,and the difference of the prediction results obtained before and after the improvement is analyzed..The experimental results show that the fault progress data generated by the simulation data generator can meet the fault prediction algorithm of the switch machine,and the research proves the real feasibility of the simulation data generator.The model prediction results and comparative experiments show that the LSTM prediction model can predict the number of actions of the switch machine with slow degeneration faults,which proves that it is suitable for the fault prediction of the switch machine,and the prediction performance of PSO-LSTM is stronger than that of the basic LSTM. |
