清华大学深圳国际研究生院王飞团队研制了一套名为"GC-Net"的地面塌陷快速预测框架，通过融合物理仿真与深度学习技术，在数据稀缺的条件下实现对塌陷过程的快速推演 1。该框架将CFD-DEM耦合模拟、网格化风险特征提取和深度学习推演三个技术环节相结合 1，可在变化的土体流失出口位置、尺寸、类型以及水流速度、方向和土体参数等多种工况下生成模拟数据进行训练 1。
为增强模型对复杂塌陷形态的识别能力，研究团队设计了位置处理模块与增强卷积神经网络融合的架构 1。该模型在南方某城市轨道交通工程建设中发生的典型地面塌陷事故中进行了独立验证，预测结果与真实塌陷形态和影响范围保持较好的吻合度 1。相关研究成果已发表在国际工程地质权威期刊《工程地质学》上，由王飞担任通讯作者，贾宇辰和何青伦为第一作者 1。该研究获得了深圳市自然科学基金等项目的资助 1。
Researchers at Tsinghua University's Shenzhen International Graduate School have created an innovative framework called "GC-Net" to predict ground collapse in urban areas 1. The system combines high-precision physics simulation with deep learning to rapidly model collapse processes even when data is limited 1.
The GC-Net framework integrates three technical components: CFD-DEM coupled simulation, gridded risk feature extraction, and deep learning-based rapid prediction 1. The team generated multi-scenario simulation data by varying soil loss exit locations, sizes, types, water flow velocity and direction, and soil parameters 1. The model also incorporates a position processing module fused with enhanced convolutional neural networks to improve its ability to identify complex collapse patterns 1.
The research has been validated through independent testing on a typical ground collapse incident that occurred during a metro construction project in a southern Chinese city, with predictions showing good alignment with the actual collapse morphology and affected area 1. The findings have been published in Engineering Geology, a leading international journal in engineering geology 1. The paper's corresponding author is Wang Fei, with Jia Yuchen and He Qinglun as first authors 1. The research received support from the Shenzhen Natural Science Foundation and other projects 1.
评论
还没有评论，欢迎留下第一条。