数据驱动的滑坡灾害预测预报方法研究

项目名称： 数据驱动的滑坡灾害预测预报方法研究

项目编号： No.61203286

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 姚为

作者单位： 中南民族大学

项目金额： 24万元

中文摘要： 滑坡等地质灾害具有突发性，并会对人类生命财产和生存环境带来巨大破坏。以当前对滑坡灾害孕灾机理的认知建立精确的滑坡运行机理模型还存在困难，而构造数据驱动的预测模型，则可以避开复杂的力学分析，为滑坡的空间预测与时间预报提供一条可行之径。本项目将先进的数据驱动建模方法引入滑坡灾害的研究中，通过对滑坡相关数据的分析，对滑坡在空间上的分布和时间上的发展变化规律进行预测。项目采用图像变化检测技术分析遥感空间数据，对滑坡风险区域进行定位；以前馈神经网络建立数据融合模型，在地面分布式测量数据的支持下，以区域地物类型信息为主要依据实现滑坡灾害的区域风险评估；基于动态的递归网络建立滑坡位移时序测量数据的预测模型，通过预测滑坡位移的变化趋势来对可能发生的滑坡灾害以及灾害发生时间进行预报。项目通过将滑坡空间预测与时间预报相结合，从而得到一套滑坡预测与预报的完整方法体系。

中文关键词： 滑坡；预测；数据驱动；神经网络；

英文摘要： Geological disasters such as landslides often happen suddenly and can bring sevre damages to human lifes and and the living environment. Based the current knowledge about the pregnancy mechanisms of landslide disaster, precise mechanism models of landslide can hardly be established, however, data-driven prediction models are much easier to construct. In these kinds of models, the complicated physical mechanisms are avoid, which makes the prediciton and forcasting of landslides a possible mission. We introduces advanced data-drive methods into the study of landslide hazard, to discover the spatial distribution patterns and time-dependent evolving trends of landslides from landslide-related data. The project contains three aspects: First, use change detection methods to analyze remote sensing images, and landlide risk can therefore be located; Then, use feed-forward neural network algorithms to establish a data fusion model, which can fuse different types of in-situ measurement data and produce spatial landslide risk map based on land cover types; Finally, employ dynamic recurrent network to establish the time series prediciont model for the in-situ landslide movement measurements, and forcast potential landslide disasters from these landslide movement predicionts. In this project, spatial prediciont and time for

英文关键词： Landslide；Prediction；Data-driven；Neural network；