An adaptive multisource fusion RTK positioning framework with spatio-temporal graph networks

This study proposes an intelligent adaptive RTK framework for multi-source heterogeneous data fusion to address GNSS positioning reliability issues in complex environments, comprising three core modules: adaptive RTK for multi-source fusion, spatio-temporal graph neural network-based ionospheric disturbance prediction, and deep ambiguity resolution network. The system integrates GNSS observations, 5G signal strength, LiDAR point clouds, and IMU data using deep learning-driven feature extraction networks to evaluate real-time confidence levels of each data source, dynamically adjust fusion strategies, and achieve quantitative mapping between environmental complexity and data source confidence, while the spatio-temporal graph neural network analyzes historical reference station data to construct ionospheric disturbance propagation maps, combining solar activity and geomagnetic data for TEC change prediction, and the deep ambiguity resolution network processes time-series double-difference observations and satellite geometry to directly output high-confidence integer ambiguity candidate sets. Experimental results demonstrate horizontal MAE of 2.8 cm and vertical MAE of 4.2 cm in urban canyon environments, 4.8 cm horizontal accuracy under 60% signal occlusion, and ionospheric disturbance prediction achieving 85.2% accuracy with 14.7-minute advance warning, realizing comprehensive breakthroughs in environmental adaptation, prediction, and efficient computation for full-scenario high-precision positioning applications.