文章摘要

文章信息/Info

Title:: Research on the Earthquake-damage Assessment of Masonry-timber Houses in Yunnan by Integrating the Principal Component Analysis with the BP Neural Network

作者:: 徐俊祖¹; 曹彦波¹; 李黎²; 张方浩¹; 徐小坤²; 赵正贤¹; (1.云南省地震局,云南昆明 650224; 2.华能澜沧江水电股份有限公司,云南昆明 650206)

Author(s):: XU Junzu¹; CAO Yanbo¹; LI Li²; ZHANG Fanghao¹; XU Xiaokun²; ZHAO Zhengxian¹; (1.Yunnan Earthquake Agency,Kunming 650224,Yunnan,China)(2.Huaneng Lancang River Hydropower INC,Kunming 650206,Yunnan,China)

Keywords:: the principal component analysis; the neural network model; masonry-timber structure; earthquake-damage assessment; Yunnan

摘要:: 针对如何选取合适的影响因素进行砖木结构房屋地震破坏合理评估的问题,提出了一种基于主成分分析与BP神经网络相融合的云南砖木结构房屋地震破坏评估方法,通过灰色关联度模型剔除对砖木结构房屋发生地震破坏影响较小的因素得到关键因子,采用主成分分析法从关键因子中提取主要成分,最后利用BP神经网络模型对处理后的主要成分进行训练,建立砖木结构房屋地震破坏比例预测模型,并利用实际震例进行验证。结果表明:本文方法相较于传统脆弱性曲线拟合方法和BP神经网络模型,其预测的砖木结构房屋地震破坏比例的预测精度更高、普适性更好。

Abstract:: There are a lot of factors affecting the earthquake damage to brick-timber houses,and selecting appropriate influencing factors is an important guarantee for an accurate and reasonable assessment of the earthquake damage to masonry-timber houses. The question is that when using the traditional methods,it is difficult to choose proper factors. In this paper,a method for assessing the earthquake damage to masonry-timber houses in Yunnan is proposed by integrating the principal component analysis and the neural network. Firstly,the less-influencing factors are eliminated through the gray correlation degree model and the key factors are obtained. Secondly,the main components from the key factors are extracted through the principal component analysis.Finally,the main components are trained through the BP neural network model,and a pre-estimating model for the earthquake-damage ratio of masonry-limber houses is established. This method is tested by using the data from the post-earthquake investigation of some historical earthquakes,and the results show that this method is more accurate and applicable for pre-estimating the earthquake damage ratio of masonry-timber houses than the traditional vulnerability curve fitting and the neural network model.