Research On Algorithms Of Data Fusion In High-speed Railway Track Deformation Monitoring

This article combines the practical high-speed railway track deformation monitoring project with the experimental scheme with which is based on track recording vehicles with 3 sensors: GNSS receivers,three-axis gyroscopes,odometers.Aiming at static monitoring and dynamic monitoring data respectively,the paper propose a number of novel data processing algorithms and use the simulated data and measured data to verify the effectiveness and advantages of the algorithm.The work mainly includes the following parts:1.As many researches indicates that the multipath error is closely related to the environment surrounding the observation station,we formulate the observationdomain single-epoch multipath as a function of the azimuth,elevation,distance of the specific satellite with respect to the station and other environment parameters.Utilizing the dataset of the first day to obtain 32 models for each GPS satellite,we apply the models to mitigate the multipath disturbance in subsequent periods.The result shows that the positioning precision is improved by about 25% in the north,east and up components.Although the effectiveness has negative correlation with the time interval between regressing day and applying day,it still exits when the time interval reaches a week ——the precision is still improved by 5%.In terms of post-processing deformation monitoring,SVR model is better than ODSF in efficiency and availability and the improvements reach 6% from ODSF.2.For the dynamic monitoring of rail vehicles in orbit,due to the particularity of the observation environment,the collected data may have gross errors,and there may be some bias in both the coefficient matrix in the observation equation and the dynamic model in the state transfer equation,so the adaptive robust total least squares filtering(ARWTLSF)algorithm is proposed and derived,which takes into account the gross error in the observation equation and the random error in the coefficient matrix and the deviation of the dynamic model.Simulation experiments show that when all these errors and variations of the model are mixed,the results of other existing filtering algorithms are so large that the state estimation values are not trustworthy.The ARWTLSF algorithm is the only algorithm available.3.If the Euler angles of the rail inspection trolley are designed as the state parameters, the elements of the rotation matrix are four operations of trigonometric functions,and the linearization is very complicated.So considering taking elements of the rotation matrix as state parameters and restricting that the rotation matrix is unit orthogonal,on the basis of the above filtering algorithm,an adaptive robust weighted least squares filtering with constraints(CARWTLSF)is proposed.Results show that its accuracy is 29%~42% higher than the traditional Kalman filtering in the sptial position.This new filtering methods can be applied to other measurement,positioning,and navigation applications.