LSSVM Enhanced Turnout Aided Train Positioning Accuracy Maintenance Method

In recent years,the development trend of train positioning system is taking GNSS as the basic means that supplemented by INS for high-precision continuous positioning of trains,so as to realize the train control system with on-board centralization and trackside equipment minimization.However,GNSS signal will be blocked due to complex environment of railway,and the system will degenerate into a separate INS that recursive calculates navigation information.The error accumulated by INS gradually increases with time,which seriously threatens the operation safety of the train.In order to solve the problems above,a method is proposed to maintain the accuracy of train positioning assisted by turnout information.The high-precision intermittently position correction of INS is realized by the turnout position assisted train positioning method,and the continuity error estimation and correction of INS are realized by the LSSVM assisted positioning method.The two methods above are combined to ensure the continuity and accuracy of train system positioning in the absence of satellite signal.The main research contents in this thesis include:(1)By learning and training the navigation data before the loss of satellite signal,the LSSVM model for INS error estimation is obtained.During the separate recursive calculation of INS,the trained model is used to predict and estimate the INS error.(2)According to the contour characteristics of the rail and the parameter characteristics of the rail at the turnout position,the characteristics of the scanning data obtained by the Li DAR are analyzed,and the methods of rail detection,rail quantity calculation,rail width calculation and rail spacing calculation are studied.With the frog as the key point of turnout detection,based on the calculation of rail parameters,a method for detecting the position of turnout frog based on parameter conditions is proposed.(3)The extension method of turnout identification based on convolutional neural network is proposed,and the network structure of CNN is built and trained to realize the detection of Li DAR scanning data of turnout frog position.(4)A method of turnout position information extraction and auxiliary positioning correction which based on GNSS / INS is constructed.The framework of LSSVM enhanced turnout aided train positioning accuracy maintenance system is completed.The experimental results show that both the turnout frog detection method based on CNN and method based on parameter conditions can detect the turnout frog effectively.By applying the LSSVM enhanced train positioning accuracy maintenance method proposed in this paper to the multisegment 10 s tunnel environment,the DRMSE in the horizontal direction is reduced from 8.77 m to 1.88 m compared with the results of traditional INS alone.In the turnout section with 40 s GNSS signal missing,by using the LSSVM enhanced turnout aided train positioning accuracy maintenance method proposed in this paper,the DRMSE in the horizontal direction is reduced from 75.65 m to 3.06 m compared with the traditional method,and the positioning performance has been significantly improved.There are 71 figures,8 tables and 66 references.