融合深度神经网络的三峡库区滑坡灾害易发性预测

Prediction of Landslide Susceptibility in Three Gorges Reservoir Area Based on Integrating Deep Neural Network

由于滑坡地质灾害数据具有多源性、异构性和非线性等特点,传统的滑坡预测手段难以充分挖掘隐藏在滑坡系统中的规律性问题。为了解决该难题,利用Stacking集成学习技术将卷积神经网络(CNN)与循环神经网络(RNN)相结合,应用于三峡库区秭归—巴东段滑坡灾害易发性预测。首先,选取20个滑坡影响因子与研究区滑坡编录数据共同构建滑坡易发性数据库;然后,将滑坡与非滑坡样本划分为训练集和测试集;最后,基于训练集和Stacking技术来融合CNN与RNN,并利用该集成方法来进行易发性预测与精度评价。结果表明,在极其有限的样本条件下,Stacking集成方法预测性能较优,其总体精度比CNN、RNN和逻辑回归高出0.87%~2.89%。

Due to the multi-source,heterogeneous and nonlinear characteristics of landslide disaster data,it is difficult for traditional landslide prediction methods to fully explore the regular problems hidden in the landslide system.In order to solve this problem,this paper uses the Stacking ensemble learning technique to combine convolutional neural networks(CNN)and recurrent neural networks(RNN),and applies them to perform landslide prediction in the Zigui-Badong section of the Three Gorges reservoir area,China.Firstly,20 landslide influencing factors are selected to build a landslide susceptibility database,along with the landslide inventory data of the study area.Then,landslide and non-landslide samples are divided into training and test sets.Finally,CNN and RNN are fused based on training set and Stacking technology,and the integration method is used to predict susceptibility and evaluate accuracy.Experimental results show that the Stacking ensemble method has better prediction performance under extremely limited sample conditions,and its overall accuracy is from 0.87%to 2.89%higher than that of CNN,RNN and logistic regression.