Civil aviation satellite navigation integrity monitoring with deep learning
基于深度学习的民航卫星导航完整性监测

摘要

GPS、伽利略、格洛纳斯和北斗等卫星导航系统提供需要高精度和高可靠性的全球定位和授时服务。监控这些系统的性能对于确保精确且强大的导航能力至关重要。传统的监测技术通过分析卫星数据来评估导航精度和服务质量。诸如精度因子（DOP）和定位误差等关键参数是根据卫星几何形状和测量值进行评估的。然而，这些方法在对复杂的性能因素进行随时间变化的建模时存在局限性。深度学习的最新进展为增强卫星导航监测提供了新的机遇。深度神经网络可以揭示海量卫星数据中隐藏的模式和动态。卷积神经网络（CNN）和长短期记忆网络（LSTM）等模型分别非常适合卫星图像处理和时间序列预测任务。深度学习方法可以融合来自全球传感器网络的多源数据，从而全面了解系统性能。神经网络可以学习原始卫星观测数据与DOP、时间漂移和定位精度等质量指标之间的复杂映射。循环模型还可以根据检测到的趋势预测未来的性能下降。关键研究重点包括收集高质量的训练数据、选择网络架构以及解释模型输出。挑战包括推广到未知缺陷和新的卫星星座。有了充足的数据和验证，深度学习可以显著改善卫星监测，从而随着系统规模的扩大和发展实现稳健的导航。总而言之，深度学习在增强卫星导航质量评估和预测方面拥有巨大的潜力。通过严谨的研究，自动化深度学习有望成为全球可靠高精度定位不可或缺的技术。这些技术的成熟及其与现有监测基础设施的整合前景光明。

Abstract

Satellite navigation systems like GPS, Galileo, GLONASS and BeiDou provide global positioning and timing services that require high accuracy and reliability. Monitoring the performance of these systems is crucial for ensuring precise and robust navigation capabilities. Traditional monitoring techniques analyze satellite data to assess navigation accuracy and service quality. Key parameters like dilution of precision (DOP) and positioning errors are evaluated based on satellite geometry and measurements. However, these methods have limitations in modeling complex performance factors over time. Recent advances in deep learning provide new opportunities to enhance satellite navigation monitoring. Deep neural networks can uncover hidden patterns and dynamics in large volumes of satellite data. Models like convolutional neural networks (CNN) and long short-term memory (LSTM) are well-suited for satellite image processing and time series forecasting tasks respectively. A deep learning approach can fuse multi-source data from global sensor networks for a comprehensive view of system performance. The neural networks can learn complex mappings between raw satellite observations and quality metrics like DOP, timing drifts and positioning accuracy. Recurrent models can also estimate future degradations based on detected trends. Key research priorities include assembling high-quality training data, selecting network architectures, and interpreting model outputs. Challenges include generalizing to unseen defects and new satellite constellations. With sufficient data and validation, deep learning can significantly improve satellite monitoring to enable robust navigation as systems scale and evolve. In summary, deep learning holds substantial promise for enhancing satellite navigation quality assessment and prediction. With rigorous research, automated deep learning could become an integral technology for reliable high-precision positioning across the globe. The outlook is positive for these techniques to mature and integrate with existing monitoring infrastructure.