Research On Train Positioning Method Under Restricted Satellite Signal Environment

In recent years,the rail transportation of China has been developing rapidly,and the requirements for train control system are getting higher and higher in order to ensure the safety of train operation,among which real-time acquisition of train position information is the prerequisite for the safe operation of train control system.At present,the train positioning method based on satellite navigation has received wide attention,but its signal is easily affected by occluding objects,which leads to a rapid decrease in positioning accuracy.To address the problem,this thesis proposes a two-pronged solution.On the one hand,in the case of good Global Positioning System signals,train position information is obtained based on combined GPS/INS positioning,and recurrent neural networks are introduced to make use of their good processing ability of time series for training and learning prediction;on the other hand,5G networks are introduced as a supplement to GPS networks,and with the help of their large bandwidth,multi antenna array,etc.,to study the wireless localization method for trains based on time delay and angle measurement parameters.The main work of this thesis is as follows:(1)Proposed a train location information prediction method based on CNNBiLSTM-Attention model.Aiming at the temporal and spatial characteristics of train trajectory,recurrent neural network was introduced,and based on this,convolutional neural network and attention mechanism are used to better focus on effective location features and improve the prediction accuracy of the model.Firstly,the pre-processed data are input into the convolutional neural network as a way to capture the spatial correlation among the train motion trajectory data;secondly,the bi-directional long and short-term memory network is used to obtain the temporal correlation among the data;finally,the attention mechanism is used to assign weights and highlight the important features to achieve better prediction performance.Experimental results show that the model proposed in this thesis improves the accuracy of train localization in the satellite signalconstrained environment.(2)For the non-tunnel area covered by 5G-R network along the railroad,various single-base station hybrid time-of-arrival/angle-of-leave/angle-of-arrival localization methods was proposed to eliminate synchronization errors in a dense multipath environment.In the one-bound scattering environment,four linear least squares,a quadratic programming and data fusion based localization algorithms was proposed to eliminate the effect of synchronization errors.Simulation results show that the proposed algorithms can effectively rectify synchronization errors,and the LLS-based localization algorithms exhibit better localization accuracy.A new dual identification algorithm(DIA)is proposed to identify multiple-bound paths in a hybrid one-bound and multiple-bound scattering environment.Compared with the existing statistical proximity test and Kmeans algorithm,the DIA algorithm is able to correctly identify multiple-bound paths,and the effectiveness of the algorithm is further demonstrated using root mean square error comparison.(3)A TDOA train location tracking method that incorporates the train formation length was proposed for the tunnel area covered by the 5G-R network along the railroad line.The method eliminates the effect of synchronization error by the estimated phase subtraction of TOA parameters obtained from the receivers at the front and rear of the train,establishes the train localization tracking model of TDOA,and adopts the existing filtering method for localization tracking.The simulation results show that the positioning model can obtain good positioning accuracy even when only one base station is used for positioning.