Research on Multi-sensor Information Fusion Positioning of Permanent Magnet Maglev Trains Based on Transformer Learning

With rapid urbanization, maglev trains have become pivotal in urban transit development owing to their high-speed operation, noise reduction, and environmental sustainability.Maglev navigation systems ensure operational safety and scheduling efficiency yet face GNSS signal interruption challenges in complex urban environments.To resolve this challenge, we propose a dual-stage fusion scheme integrating Transformer-based neural network prediction with Kalman fusion.Integrating multi-sensor fusion with Transformer neural networks enhances maglev train positioning under GNSS interruption scenarios.  This approach utilizes historical GNSS, INS and other sensor data to learn spatiotemporal correlations and motion trends, enabling predictive positioning during signal outages.   Experiments demonstrate that the Transformer-Kalman fusion model achieves  fusion trajectories are closer to the real values, yielding lower MAE, MAPE, and RMSE errors than alternative structures.   Compared with the Extended Kalman Filter (EKF) and Adaptive Kalman Filter (AKF) algorithms in different scenarios, the neural network model based on Transformer-Kalman fusion predicts the two-stage fusion scheme with significant advantages in terms of stability and accuracy, enhancing the robustness of the positioning system.   Provide key technical support for the safe operation of maglev trains.