Detection of GNSS NLOS signals in urban environments via stacking ensemble learning and signal features

In urban environments, Global Navigation Satellite System (GNSS) signals are highly susceptible to obstructions from tall buildings, leading to Non-Line-of-Sight (NLOS) errors, and severe positioning degradation. Machine Learning (ML)-based NLOS detection has gained significant attention, due to its high accuracy and the advantage of requiring no hardware modifications. However, the existing studies predominantly rely on single-model architectures, which often suffer from poor generalization, and a tendency to converge to local optima. To overcome these limitations, this study proposes a GNSS NLOS detection method based on a two-layer Stacking Ensemble Learning (SEL) model and five key GNSS signal features. A comprehensive weighting model is then applied to correct NLOS-induced pseudorange errors. Results show that the SEL model, utilizing our selected GNSS signal feature set for NLOS detection, achieves a 13.1% improvement in classification accuracy, compared with the traditional feature sets. This improvement is primarily attributed to the more precise and comprehensive selection of features, which collectively consider satellite geometry, signal strength variations, dynamic characteristics, and pseudorange errors. Moreover, the SEL model integrates multiple heterogeneous base models, demonstrating superior generalization capability and higher detection accuracy, making it particularly well suited for GNSS observation data processing in complex urban areas, encompassing multiple environments. Specifically, it achieves NLOS classification accuracies of 94.5% and 93.1% in low-speed and high-speed dynamic scenarios, respectively. Furthermore, the SEL model exhibits lower positioning errors and enhanced robustness. Compared with single base models, it reduces horizontal and vertical positioning errors by 29.7% and 25.7% in low-speed dynamic scenarios, respectively, while by 26.3% and 25.9% in high-speed dynamic scenarios. Our proposed method requires no additional hardware modifications to low-cost receivers while achieving high positioning accuracy and reliability, offering a new approach for continuous high-precision navigation in smart city applications.