Landslide movement calculation method for multiple catchment areas based on homologous physical information neural network
By using a pre-training-fine-tuning method of homogeneous physical information neural networks, combined with depth-averaged shallow water control equations and the finite difference method, the high cost and low efficiency of landslide motion calculation in multi-catchment areas are solved, enabling rapid and accurate prediction and efficient assessment of landslide motion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI JIAOTONG UNIV
- Filing Date
- 2026-05-21
- Publication Date
- 2026-07-17
AI Technical Summary
Existing landslide movement prediction methods are computationally expensive and time-consuming at the regional scale of multiple catchment areas, and purely data-driven models lack physical consistency and generalization ability, making it difficult to meet engineering reliability requirements.
A pre-training-fine-tuning method based on a neural network with homogeneous physical information is adopted, combined with depth-averaged shallow water control equations and the finite difference method, to construct a multi-catchment landslide motion calculation model with shared physical knowledge. Through pre-training, size scaling, adaptive weight balancing, and flow depth weight sampling, the rapid and accurate calculation of landslide motion is achieved.
It improves the efficiency and reliability of landslide motion calculation, reduces the difficulty of iterative optimization, enhances the training accuracy and generalization ability of the model, and realizes rapid and accurate prediction of landslide motion in multi-catchment areas.
Smart Images
Figure CN122413974A_ABST