A Novel Data-driven Adaptive Robust Filter based on TCN-ACKF Method for Autonomous Train Localization System

Satellite positioning technology plays a significant role in the next-generation train operation control system, which contributes to the establishment of an autonomous railway transportation system with reduced operation and maintenance costs. The quality of GNSS (Global Navigation Satellite System) satellite signal is not always sufficient for accurate and reliable positioning in complex railway environments. Uncertain measurement noise and faults may result in significant performance degradation. To address these issues, an adaptive robust Cubature Kalman filter (ACKF) method is proposed based on the Temporal Convolutional Network (TCN). An adaptive measurement noise covariance update module and a fault detection module are designed under the novel TCN-ACKF scheme. The adaptive measurement noise covariance is calculated with the pseudo-range error predicted by the TCN. The fault detection module is employed to determine whether there is a fault and carry out the specific measurement compensation when fault occurrence is identified. Simulations are conducted to evaluate the proposed method under a challenging GNSS environment via field data collected on the Lhasa-Rikaze Railway. Different scenarios, with both the step fault and time-varying noise, are investigated to illustrate the performance of the proposed filter. Experimental results demonstrate that the proposed method achieves robust state estimation against uncertain measurement noise and fault, significantly enhancing the positioning accuracy and robustness of the autonomous train positioning system.