Abnormal Detection Of Turnout Operation State Based On Support Vector Domain Description

Turnout as the key equipment of railway line connection,once the failure occurs,it will affect the running efficiency slightly,or seriously endanger the running safety,leading to train derailment.For a long time,the maintenance mode of railway turnout equipment in our country mainly adopts periodic maintenance,which is easy to cause excessive maintenance and insufficient maintenance.It also uses periodic browsing and analysis of monitoring data to achieve "state maintenance".This mode of "state maintenance"needs huge energy of staff,and may cause missed inspection.Therefore,the purpose of this paper is to reduce the labor intensity of staff and improve labor efficiency by using automatic detection method.This paper uses power data to detect abnormal force state of traction point during turnout operation.Based on the analysis of the actual power data of turnout in Changsha Electric Depot,it is found that the main problem is that there are more normal samples and fewer abnormal samples in the actual data.In this paper,support vector domain description algorithm(SVDD)is proposed to model the turnout anomaly detection,which can solve the problem of large number of actual normal samples.Two classes of samples with interval support vector domain description algorithm(2C-SVDD)is proposed to improve the robustness of the model.A parameter optimization method based on grid ant colony is proposed to find the optimal parameter model of the algorithm.An on-line learning-based support vector domain description(OSVDD)algorithm is used to model turnout anomaly detection without historical data.The main work of this paper is as follows.(1)Through the analysis of the action process of ZYJ7 hydraulic switch machine,this paper finds out the main reason of abnormal traction force during the operation of the switch and divides the power curve of the switch into five stages.Through the analysis of the control circuit and power acquisition principle of the switch,the data needed to establish the abnormal detection model of the switch are determined.(2)In this paper,an overall method of turnout anomaly detection system is proposed,and the principles of SVDD,2C-SVDD and OSVDD algorithms are introduced in detail.Principal Component Analysis(PCA)was used to reduce the dimension of the extracted time-domain features,and the parameter optimization method of grid ant colony algorithm was used to find the optimal parameters and the optimal input dimension of the model.The experimental results show that when the eigenvector is reduced to 2-D5 the accuracy of the model is the highest.Experiments show that the performance of 2C-SVDD is better than that of SVDD,and that of OSVDD is better than that of SVDD.(3)In this paper,the general design principle and framework of turnout anomaly detection system software are presented.The software is modularized by UML,and class diagram as well as use case diagram of software are drawn.Then the specific functions of each module are designed,such as user management module,interface display module and database management module.Finally,the software is programmed by C#.Finally,the abnormal detection system is tested by using the data of turnout in Changsha Station.The accuracy of abnormal detection in the three stages of its unlocking,conversion and locking is all over 97%,which shows that the abnormal detection method has good performance and meets the needs of maintenance department.