摘要
为研究城市地铁沿线老旧房屋普遍存在结构安全问题,基于机器学习模型,选取房屋年份、楼层、面积等11个属性构建预警指标体系,采用SMOTE过采样、独热编码等方法解决样本离散化、不均衡的问题;利用KNN、Bayes、Logistic、SVM 4种机器学习模型对房屋结构安全数据学习并测试,综合应用Accuracy、F1、AP、AUC等指标比较预警模型性能。结果表明:以某市地铁1、2号线沿线大于20 a的2431栋老旧房屋为例,得到Logistic和SVM的预警精度较高,影响地铁沿线老旧房屋安全现状的主要因素为房屋改造情况、基础、结构类型和设计情况,验证了模型效果。
In order to study the structural safety problems commonly existing in the old buildings along the urban metro lines,based on the machine learning model,the advanced and accurate early warning for the safety of old buildings along the metro lines was realized.11 attributes such as building year,floor,area and others were selected to build an early warning index system,and the SMOTE oversampling and one-hot encoding methods were used to solve the problems of sample discretization and imbalance.Four machine learning models including KNN,Native Bayes,Logistic and SVM were used to learn and test the structural safety data of buildings,and the Accuracy,F1 score,AP Score,AUC score and other indexes were comprehensively applied to compare the performance of the early warning model.Taking 2431 old buildings over 20 years along metro lines 1 and 2 in Gulou District of Fuzhou as examples,the results showed that the early warning accuracies of Logistic and SVM were higher.The main factors affecting the safety status of old buildings along the metro lines were the building transformation condition,foundation,structure type and design situation,which verified the model effect.
作者
段在鹏
李帆
邱少辉
俞思雅
张一洋
DUAN Zaipeng;LI Fan;QIU Shaohui;YU Siya;ZHANG Yiyang(College of Environment & Safety Engineering,Fuzhou University,Fuzhou Fujian 350108,China;Chinalco Ruimin Co.,Ltd.,Fuzhou Fujian 350108,China)
出处
《中国安全生产科学技术》
CAS
CSCD
北大核心
2022年第3期162-167,共6页
Journal of Safety Science and Technology
基金
国家社会科学基金项目(17CGL049)。
