Research On Train Positioning Method Based On Multi-source Information Fusion

With the rapid development of railway transportation in China,new train control systems,efficient train scheduling,assisted driving,and safety monitoring have put forward higher requirements for stable and reliable positioning of trains.Multi source information fusion technology can effectively improve train positioning performance,thereby better meeting the high-quality positioning needs of trains.This article focuses on the stable and reliable positioning of trains in complex environments.Focusing on multi-source information fusion,research is conducted on the combination of prior information constraints and multi-source information fusion positioning methods for trains under various complex conditions.The theoretical framework of multi-source information fusion positioning for trains in complex environments is expanded,providing theoretical and technical support for promoting the application of multi-source information fusion positioning technology in China’s rail transit.The main research content is as follows:(1)In order to fully utilize the prior information of train positioning to improve its positioning performance,an adaptive H_∞filtering fusion positioning method based on the combination constraint of prior information of trains is proposed.Firstly,based on analyzing the characteristics of train track information,a high-precision digital track generation method is studied,and then a train track constraint model is established;Combining the characteristics of train motion,construct a combination constraint of train prior information;The proposed method utilizes fuzzy adaptive adjustment of the acceleration limit value of the train positioning system to improve the adaptive ability of the H_∞filter.With the addition of prior information combination constraints,it effectively improves the positioning accuracy of the train positioning system.Compared with traditional filtering algorithms,the positioning accuracy is improved by more than 40%.(2)A train visual inertial navigation fusion positioning method based on wavelet cascaded filtering is proposed to address the poor positioning performance of train positioning systems in satellite navigation system failure environments.The train visual odometer based on improved sparse optical flow method is constructed by using visual sensors;After preprocessing the image using the Haar wavelet transform method,the kilometer markers in the image are detected and digitally segmented,and then a simplified Le Net-5 network is used to achieve rapid recognition of kilometer markers;Combining visual odometers,onboard odometers,inertial navigation systems,and kilometer marker position recognition,the data is denoised using adaptive threshold wavelet,and the train position is estimated through adaptive H∞filtering.This method can achieve visual inertial navigation fusion positioning in the case of complete failure of satellite navigation systems,ensuring effective train positioning in the event of satellite loss of lock,and expanding train positioning methods in extreme cases.(3)On the basis of in-depth research on multi-sensor reliability evaluation methods,a dual factor federated filtering train fusion localization method based on sliding window confidence entropy is proposed for redundant structures.Firstly,study the different redundant structures and reliability in the train positioning system,and analyze the weighted fusion method for redundant structures;By measuring information precision and certainty,combined with information consistency,a method for evaluating information reliability based on sliding window confidence entropy is proposed;Using the robustness factor based on Mahalanobis distance and the adaptive factor based on prediction residual,a two-stage hybrid federated filter with parallel structure is designed.In the main filter,information fusion is carried out through sliding window information entropy.The proposed positioning method is suitable for redundant positioning structures and can effectively control the impact of interference and system model deviation on system state estimation,improving the fault tolerance and reliability of the train positioning system.(4)In order to reduce the positioning instability caused by frequent switching of trains during continuous positioning inside and outside the station,confidence entropy is introduced into the interactive multiple model algorithm to achieve stable and reliable automatic switching inside and outside the station.Firstly,compare and analyze the configuration and performance of ultra wideband base stations in the station,and construct a combined positioning model of ultra wideband and inertial navigation system;By establishing the relationship between state noise and train state,the covariance of system state noise during the positioning process is adaptively adjusted,and then the transition probability matrix is modified based on confidence entropy to improve the model jump performance in interactive multiple models.In the main filter,the estimation probabilities of each model and sliding window confidence entropy are weighted and fused to the estimation values of the sub filter.The proposed method can achieve automatic switching of positioning models during the continuous positioning process of trains inside and outside the station,improving the stability and accuracy of train positioning,and the positioning accuracy can be improved by about19.8%.