The performance of integrated navigation using global navigation satellite systems (GNSS) and inertial navigation systems (INS) in challenging urban environments is highly dependent on the quality of GNSS observations, particularly under multipath and non-line-of-sight conditions. Conventional weighting strategies based on predefined empirical relationships with satellite elevation angle or signal quality are limited in characterizing the nonlinear and time-varying characteristics of GNSS measurement errors. To address this issue, this study proposes a deep learning-based adaptive weighting method for tightly coupled GNSS/INS integration. Satellite elevation angle, signal-to-noise ratio (SNR), observation residuals, and normalized observation residuals are jointly utilized to characterize observation quality, and a neural network is employed to estimate satellite-specific measurement noise parameters for constructing the measurement noise covariance matrix. Since ground-truth measurement noise is difficult to obtain directly, the network is optimized using position, velocity, and attitude errors as the training objective. Vehicle-based experiments were conducted in representative environments, including open-sky areas, tree-lined roads, urban street canyons, and overpass sections. The proposed method achieves east, north, and up position root mean square errors (RMSEs) of 0.38, 0.75, and 1.14 m along route A, corresponding to reductions of 15.6%, 17.6%, and 17.4% compared with the combined elevation angle and SNR weighting strategy. Along route B, the corresponding RMSEs are 0.56, 0.71, and 1.18 m, representing reductions of 25.3%, 43.2%, and 34.1%, respectively. Consistent improvements are also achieved in velocity and attitude estimation. These results demonstrate that the proposed adaptive weighting strategy effectively improves the accuracy and stability of tightly coupled GNSS/INS integration under challenging observation conditions.

Accurate and continuous onboard real-time orbit determination (RTOD) is essential for Low Earth Orbit (LEO) satellites, but temporary interruptions of onboard Global Navigation Satellite Systems (GNSS) observations can degrade orbit continuity and delay filter re-convergence. Although predicted-orbit constraints can accelerate RTOD re-convergence after GNSS outages, their effectiveness largely depends on a realistic stochastic representation of the accumulated prediction uncertainty. Here, we develop a prior-state covariance model to characterize the time-dependent prediction uncertainty and provide uncertainty-consistent predicted-orbit constraints after observation recovery. Using onboard GNSS data from LuTan-1A as the primary dataset, orbit prediction residuals were analyzed through marginal-distribution analysis, envelope statistics, and a Mahalanobis-distance-based goodness-of-fit test. Results indicate that prediction errors are consistent with an approximate Gaussian model over the investigated prediction interval, supporting construction of a time-continuous prior-state covariance matrix. The proposed model was subsequently evaluated using a temporally independent LT-1A dataset. For GNSS interruptions of 15–90 min, the proposed model achieved average re-convergence times of 0.59–1.53 min, compared with 1.41–3.90 min for the state transition matrix (STM)-propagated covariance strategy and 9.18–19.12 min for the loose-prior strategy, while maintaining post-convergence 3D position RMS errors of 12.26–13.83 cm. Relative to STM propagation, average re-convergence time decreased from 1.41 to 0.59 min for the 15 min interruption and from 3.12 to 1.46 min for the 60 min interruption. Additional validation using GRACE-FO C, Sentinel-3A, and Sentinel-6A demonstrated applicability across different orbital and GNSS observation conditions. Overall, these results demonstrate that realistic characterization of prediction uncertainty can substantially improve RTOD re-convergence while preserving post-convergence orbit accuracy.