爆破振动诱发民房结构损伤识别的随机森林模型  

Random forest model for identification of residential structure damage induced by blast vibration

作　　者：方前程[1] 商丽[1] 商拥辉[1,2] 宋译[3] 

机构地区：[1]黄淮学院建筑工程学院,河南驻马店463000 [2]中南大学土木工程学院,湖南长沙410075 [3]湖南科技大学能源与安全工程学院,湖南湘潭411201

出　　处：《爆炸与冲击》2017年第6期939-945,共7页Explosion and Shock Waves

基　　金：国家自然科学基金项目(11072072)

摘　　要：为快速、准确地评价爆破振动诱发民房结构损伤效应,借鉴随机森林理论并结合工程实际,建立露采爆破振动诱发民房结构损伤识别的随机森林模型;综合考虑爆破参数、爆破振动特征参量及房屋结构动力特性等因素,选取质点峰值振动速度、主频率、主频率持续时间、段药量、爆心距、施工质量参数、场地条件参数、屋盖形式参数、砖墙面积率、民房高度、灰缝强度和圈梁构造柱参数等12个影响因素作为模型输入,将砖混结构建筑物的损害等级作为模型输出;基于多分类器集成的思想,以108组爆破振动实测数据作为学习样本进行训练,建模过程中由多个决策树集成随机森林、用投票的方式实现对民房结构损伤有效识别;用12组现场数据验证模型的有效性;在对样本分类的同时,计算预测变量的重要性值,发现质点峰值振动速度为最重要的评价指标,其后依次为爆心距,主频率持续时间,主频率,圈梁构造柱参数,灰缝强度,屋盖形式参数,民房高度,段药量,施工质量参数,砖墙面积率和场地条件参数。研究结果表明:随机森林模型预测结果学习样本准确度是87.97%,而测试集准确度是91.67%,与实际情况吻合较好,预测精度较高。In this work,aiming to the prediction speed and accuracy,we established a random forest model for residential structure damage induced by blast vibration identification on the basis of the random forest（ RF） theory. Twelve indexes,i.e. peak particle velocity,dominant frequency,dominant frequency duration,maximum charge per delay,distance,gray joints intensity,rate of brick walls,height of housing,roof structures parameter,beam-column frames parameter,quality parameter of construction and site conditions parameters,were considered as the criterion indices for this kind of damage in the proposed model based on the of analysis of the characteristic parameters of blasting vibration and dynamic characteristics of the housing structure. 108 sets of vibration measured data were investigated to create an RF classifier. RF was a combination of tree predictors,and variable importance was measured by gini importance parameter when the forest grows. A random tree was a combination of decision trees,and each tree is generated depending on the values of random vectors sampled independently,with the same distribution for all trees in the forest. The Gini importance value shows that the peak particle velocity is the most important discrimination indicator,followed by the distance,the dominant frequency duration,the dominant frequency,the beam-column frames parameter,the gray joints intensity,the roof structures parameter,the height of housing,the maximum charge per delay,the quality parameter of construction,the rate of brick walls and the site conditions parameters. Another twelve groups of residential structure damage instances were tested as forecast samples,and the predicted results were identical with the actual situation. Engineering practices indicate that the accuracy of the RF method of learning samples is 87.97%,and the accuracy of the test samples is 91.7%,effectively verifying and supplementing the existing methods for evaluating residential structure damage induced by blast vibration.

关 键 词：爆破振动 民房结构损伤 随机森林 质点峰值振动速度 预测 

分 类 号：O381[理学—流体力学]