Research On Train Positioning Method For Urban Rail Transit Based On Multisensor Fusion

The acquisition of accurate train position is an important foundation of train control system.During the running of the train in the section,due to external interference and vehicle body fluctuation,uncertainty of Eurobalise absolute position correction exists;Due to the impact of continuous wheel wear and non ideal circular motion,deviation between the software configured wheel diameter and the actual wheel diameter cause uncertainty accumulation of distance measurement by the axle speed sensor.During the process of parking in station by jump,distance measurement of radar is inaccurate at low speeds and the axle speed sensor is easily to be disturbed by external environment at low speeds.Multiple starts and stops can cause accumulation of position uncertainty.Train tracking interval can be shorten by reducing the accumulation of positioning uncertainty during section running,and closed-loop control feedback and stop accuracy statistics during the parking process can be achieved by continuous correction of absolute positions within the local area of the platform.This conforms to the development direction of fully automatic operation systems and train autonomous control system based on vehicle to vehicle communication which aims to improve the level of automation and intelligence.The theory and method of train positioning are studied in the thesis,including absolute position correction,relative position correction,wheel diameter calibration,multisensor fusion and other key technologies.The main research contents and innovative work include:(1)In order to reduce the absolute position correction uncertainty of Eurobalise,the composition and internal information flow of Eurobalise position correction system are studied.It is found that the effective electromagnetic action time error and velocity measurement error have a great impact on the position uncertainty.For the issue of electromagnetic action time,the uplink electromagnetic induction dynamic model of Eurobalise system is established.The model using quadrilateral analog balise transmit antenna and octagonal analog Balise Transmission Module(BTM)receive antenna,introducing train movement into magnetic flux change,analyze the influence of dynamic electromotive force and induced electromotive force on the electromagnetic action distance.The compliance of the field measured Eurobalise correction data with the electromagnetic induction dynamic model is verified.The support vector machine abnormal classification model is obtained by machine learning algorithm and position correction training dataset which is created by marking abnormal data manually.With this classification model,abnormal Eurobalise position correction data can be classified online,and will not be used for the wheel diameter calibration.(2)In order to improve the accuracy of wheel diameter calibration,the basic principles of the existing manual and automatic wheel diameter measurement methods are analyzed,and a wheel diameter calibration scheme is proposed to derive the real-time wheel diameter calibration based on measurement deviation of Eurobalise interval on the whole line.In order to reduce the proportion of Eurobalise installation deviation and position correction process speed and time calculation deviation in the measurement deviation of Eurobalise interval,the Kalman filter state equation and update equation for the measurement deviation of Eurobalise interval are constructed using the ratio of adjacent Eurobalise interval as the state transition coefficient.The wheel diameter deviation is inversely calculated based on the optimal posterior estimation value of the Eurobalise interval measurement deviation obtained through iterative calculation.Due to the unknown measure noise variance,an adaptive algorithm using Support Vector Regression(SVR)based on radial basis function(RBF)kernels is proposed to estimate the measure noise variance of the Eurobalise interval measurement deviation.This model is used to process field data,and the performance differences between standard Kalman filter and adaptive Kalman filter based on Support Vector Regression are compared and analyzed.(3)Aiming at the problem of uncertainty accumulation in station parking by jump,a method of continuous correction of train absolute position in local area using laser ranging sensors to measure the trackside installation height difference matrix is proposed.The design methods of positioning matrix and vehicle ranging sensor group are discussed,the applicability selection schemes of different laser point measurement technologies are analyzed.In order to deal with the interference of vehicle vibration on laser ranging,a positioning estimation method based on static binary Bayesian filtering technology is proposed.The accuracy,range and robustness of the positioning matrix system are analyzed.Based on this positioning matrix system,the cumulative change of positioning uncertainty during the final stage of automatic train operation(ATO)system station parking is tested and verified.(4)Aiming at the problem of improving positioning accuracy through multisensor fusion,the characteristics of different sensor fusion architectures are analyzed,a hybrid train positioning system multisensor fusion architecture is designed to meet the needs of absolute position correction and relative position correction.The information fusion of the Eurobalise,the positioning matrix,the axle speed sensor and the radar sensor installed at both ends of the train,is realized through the network,and the strategy of data fusion is proposed.For the optimization of detection and fusion in fusion system,the theoretical relationship between Bayes risk and the decision rules of fusion center and sensor is analyzed.Aiming at the centralized parallel sensor architecture used by the relative positioning sensors,taking the radar abnormal judgment as an example,a fusion determination mechanism which contains three levels,namely,fusion center,velocity measurement distribution system,and sensor,is designed using D-S evidence theory.(5)The indoor simulation test environment is built based on Hardware-in-the-Loop(HIL)technology,and the virtual controlled equipment model is used to verify and test position function of real automatic train protection(ATP)system,ATO system,and other on-board controllers.The simulation test environment is used to verify Eurobalise position correction model,wheel diameter calibration model,positioning matrix model and multisensor fusion technology.Based on the actual operation data of Guangzhou Metro Lines 18 and 22,it is verified that the theories and methods studied in this thesis can improve the position accuracy and automation level of the train control system,and can improve the robustness of the system.