A BDS/LTE-R Enhanced Data Fusion Positioning Algorithm Based on Deep Learning

Abstract

Integrated positioning methods in the high-velocity train control system require auxiliary equipment leading to more construction and maintenance costs. This paper proposes a BDS/LTE-R (Beidou Navigation Satellite System/Long Term Evolution-Railway) integrated positioning system based on deep learning and establishes a 7L-CNNSeven-layer convolutional neural network) an enhanced model of BDS/LTE-R data fusion positioning. Firstly, the positioning principles of the BDS and LTE-R system are analyzed to construct a data space second order autocorrelation matrix of the results from each single positioning system, which serves as the input of the 7L-CNN model. The positioning data are output after the depth feature extraction and feature fusion. In a test based on field data, the 7L-CNN model obtains fusion results with the second-order autocorrelation matrix of positioning data space as the input. Compared with the results obtained from the early fusion algorithm and CNN (convolutional neural network) fusion positioning model with the input of the original positioning data, the 7L-CNN enhanced algorithm can bring better convergence accuracy for both solving velocity and positioning results according to the velocity and position errors in the east and north directions. When a satellite is out of the lock, the 7L-CNN algorithm also has a good correction effect on the single LTE-R positioning, which can meet the requirements for high-precision and continuous real-time positioning of a train.

Keywords

• traffic information engineering and control
• integrated train positioning
• deep learning; enhanced data fusion positioning
• positioning accuracy