Research On Algorithms Of Data Fusion In High-speed Railway Track Measurement System

A project named ”GNSS/INS combined high-speed railway track irregularity measurement technology and application” was proposed to solve the problem of low efficiency and high cost in existing high speed railway track measurement solutions byavoiding dependence on CPⅢ control network.The system consists a GNSS base station network and a railway measurement trolley,equipped with IMU,odometer,GNSS receiver and other sensors.Static measurement was adopted to meat the accuracy requirement in high-speed railway track measurement,which causes a giant decrease in efficiency.As a part of the project,this research aims to solving the dependence of absolute coordinate accuracy on static measurement and achieve full dynamic measurement both in absolute and relative measurement,which may greatly improve the efficiency.First,the main sensors,IMU,GNSS receiver and odometer’s error was analyzed and modeled,which is the basis of data fusion models.Then,the data fusion models for Kalman filter of high-speed railway track irregularity measurement system was established and solved.To solve the shortless of Kalman filter that the information cann’t passed backward,the Kalman-RTS smoother was used.Then,in order to solve the problem of the relative accuracy reducing,graph optimization was used in the system data fusion.At last,in order to improve the absolute coordinate accuracy,the GNSS double difference residual was modeled and a tight combined high-speed railway track irregularity measurement system data fusion model was established and solved by graph optimization.The calculation results are as follows:For absolute measurement,the accuracy of the original static measurement plane co-ordinate is about 2mm,and middle section recursive coordinate accuracy is related to the stop interval.The accuracy of Kalman filtering is poor,with average deviation of 2.9mm,and the maximum deviation is 24.5mm.By using Kalman-RTS smooth,The absolute co-ordinate accuracy was improved with a average deviation of 1.6mm,and the maximum deviation of 5.4mm.As for graph optimization,the result is similar with Kalman-RTS smoother,where the maximum value of the plane coordinate deviation is 4.7mm in tight combined graph optimization.For relative measurement,in original solution,mean deviation of relative plane coor-dinate is 0.2mm,and the maximum deviation is 1.5mm.Kalman filter’s results are poor,the accuracy is lower than the original solution,with an average deviation of 2mm and a maximum deviation of 23.8mm,which is due to the inability of backward propagation of information.The accuracy of Kalman-RTS is slightly lower than original solution,with an average deviation of 0.2mm and a maximum deviation of 3.6mm,in which the maximum deviation has a greater disadvantage compared with the original solution.The accuracy of loose combined graph optimization was improved with an average deviation of 0.3mm and a maximum deviation of 2mm.The accuracy of the tight combination graph opti-mization is significantly improved,with an average deviation of 0.1 mm and a maximum deviation of 0.6 mm.It has been verified by experiments that through the tight combined graph optimiza-tion diagrams,the accuracy of absolute and relative measurement is closed to the original solution in full dynamic measurement,which can improve efficiency by more than 700%.