青藏铁路跨越约550 km的多年冻土区，铁路安全运营的关键取决于冻土层稳定性。选择铁路冻土段沿线2022年10月至2024年12月的Sentinel-1A升降轨数据，利用SBAS-InSAR技术反演形变特征并分析驱动因子，融合多模型优势构建预测模型并分析。结果表明：①铁路沿线视线（Line of Sight）向升降轨形变速率分别为-35~42 mm/a和-36.5~30 mm/a；在忽略南北向地表形变前提下，垂直和东西向平均速率分别为-1.62 mm/a和2.31 mm/a，表现为下沉和向东运动。②基于“热力—水文—地形”耦合分析，各因子与形变速率均呈正相关，地表温度与活动层厚度为主控因素，贡献度为62.4%；坡度（21.4%）决定形变的方向；气温（16.2%）通过地表能量转换间接驱动。气温和地温升高引发活动层增厚导致融沉，在坡度作用下转化为坡向蠕滑位移。③基于分层阈值法识别出3类形变强度区，其中强形变区威胁路基稳定，中形变区存在冻土蠕滑风险，弱形变区形变受河流作用控制。特征点时序分析揭示了冻土区典型的季节性冻胀融沉规律及水热与沉积叠加影响。④预测分析了P1~P7共7个特征点的形变趋势。结果表明，混合模型在各特征点上的预测精度均优于单一模型，周期性特征点未来6个月的形变趋势与实际形变高度吻合，在所选特征点上表现出良好的预测性能。研究结果验证了该方法在冻土区铁路沿线形变监测中的应用潜力，可为铁路稳定性评估与路基维护提供参考。

The Qinghai-Xizang Railway spans more than 550 kilometers of permafrost regions, and the key to the safe operation of the railway depends on the stability of the permafrost layer. Therefore, this paper selects Sentinel-1A ascending and descending orbit data along the railway permafrost section from October 2022 to December 2024, uses the SBAS-InSAR technique to invert deformation characteristics and analyze the driving factors; combines the advantages of multiple models to build a prediction model and conducts an analysis. The results show: ① The Line of Sight (LOS) deformation rates along the railway for ascending and descending orbits are -35~42 mm/a and -36.5~30 mm/a, respectively; neglecting north-south surface deformation, the average vertical and east-west rates are -1.62 mm/a and 2.31 mm/a, showing subsidence and eastward movement; ② Based on the “thermal-hydrological-topographic” coupling analysis, all factors are positively correlated with the deformation rate. Surface temperature and active layer thickness are the main controlling factors, contributing 62.4%. Slope (21.4%) determines the direction of deformation. Temperature (16.2%) indirectly drives through surface energy conversion. The increase in temperature and ground temperature leads to the thickening of the active layer, causing thaw settlement, which is transformed into slope creep displacement under the influence of slope. ③ Three types of deformation intensity zones were identified using the stratified threshold method. Among them, the strong deformation zone threatens the stability of the embankment, the medium deformation zone has the risk of permafrost creep, and the weak deformation zone’s deformation is controlled by river action. The time series analysis of characteristic points reveals the typical seasonal freeze-thaw settlement pattern in the permafrost region and the superimposed influence of water, heat, and sedimentation. ④ The deformation trends of seven characteristic points P1 to P7 were predicted and analyzed. The results show that the hybrid model has higher prediction accuracy than the single model at each characteristic point. The deformation trends of periodic characteristic points in the next six months are highly consistent with the actual deformation, demonstrating good prediction performance at the selected characteristic points. The research results verify the application potential of this method in deformation monitoring along the railway in permafrost regions and can provide a reference for railway stability assessment and embankment maintenance.