BP神经网络在一维降雨入渗深度的仿真应用  

Application of BP Neural Network in the Prediction of One-dimensional Rainfall Infiltration

作　　者：覃茂森 程圣国[1] 刘朝[1] 赖坡 

机构地区：[1]三峡库区地质灾害教育部重点实验室,湖北宜昌443002

出　　处：《水力发电》2017年第6期39-42,共4页Water Power

基　　金：三峡库区地质灾害教育部重点实验室开放基金项目(2015KDZ01);三峡大学研究生创新基金项目(SDYC2016022)

摘　　要：降雨引起的土体滑坡中,绝大多数滑动面都在边坡土体的最大入渗深度范围之内,快速识别降雨入渗深度对滑坡前期的预测预报具有重要的意义。以室内一维降雨入渗试验为基础,抽取表征降雨入渗深度的关键特征值,采用主成分分析法,将影响降雨入渗深度的5个特征值综合成2个主成分,基于主成分分析建立BP神经网络仿真预测模型。预测结果与实际入渗深度最大误差值仅为4.69%,说明网络性能良好。In the landslides caused by rainfall, most of the sliding surfaces are within the maximum infiltration depth of slope soil. The rapid identification of rainfall infiltration depth has great significance to the early prediction and warning of landslide. Based on one-dimensional rainfall infiltration experiment, the key characteristic values of rainfall infiltration depth are extracted, and the principal component analysis method is used to analyze the influential factors of rainfall infiltration depth. The five eigenvalues are integrated into two principal components, and the BP neural network prediction model is established based on principal component analysis. The results show that the maximum error of prediction is 4.69%. That means the network performance is good.

关 键 词：BP神经网络 降雨 入渗深度 主成分分析 

分 类 号：TU411.4[建筑科学—岩土工程]