Enhancing the reliability of displacement prediction of reservoir landslide through a physics-informed data augmentation framework
编号:45
访问权限:仅限参会人
更新:2025-07-22 15:26:44 浏览:54次
口头报告
摘要
Accurately predicting landslide displacement remains challenging due to complex movement mechanisms and large uncertainties. Here, we develop a physics-informed data augmentation framework to tackle this challenge. By incorporating external drivers, including rainfall and reservoir water level, we establish quantitative relationships between key aging parameters (e.g., creep and strength parameters) and target indicators (e.g., responses of monitoring point (RMS)) using surrogate models. We use time-series monitoring data to sequentially update the probability distribution of aging parameters. The posterior distributions are used to adaptively and dynamically predict landslide displacement. We find that the temporal evolution of creep parameters closely resembles the observed displacement, exhibiting both a periodic pattern and a long-term growth trend, while the strength parameters show a consistent deterioration over time. These findings offer a mechanistic interpretation of the step-like deformation behavior. The integration of movement mechanisms and observed data not only reveals the evolutionary progress of aging parameters but also improves the reliability of prediction by effectively reducing uncertainties.
关键词
Reservoir landslide; Displacement prediction; Sequential Bayesian updating; Parameter evolution; Physics-informed data augmentation
稿件作者
Longhou Gan
Tongji University
Ming Peng
Tongji University
发表评论