基于Stacking模型融合的深基坑地面沉降预测  

Prediction of Ground Settlement Around Deep Foundation Pit Based on Stacking Model Fusion

作　　者：秦胜伍[1] 张延庆 张领帅[1] 苗强 程秋实 苏刚 孙镜博 Qin Shengwu;Zhang Yanqing;Zhang Lingshuai;Miao Qiang;Cheng Qiushi;Su Gang;Sun Jingbo(College of Construction Engineering,Jilin University,Changchun 130026,China)

机构地区：[1]吉林大学建设工程学院,长春130026

出　　处：《吉林大学学报（地球科学版）》2021年第5期1316-1323,共8页Journal of Jilin University：Earth Science Edition

基　　金：国家自然科学基金项目(41977221);吉林省科技发展计划项目(20190303103SF)。

摘　　要：为了提高机器学习对深基坑地面沉降的预测能力,本文提出了一种基于Stacking集成学习方式的多模型融合的地面沉降预测方法,并以深圳某深基坑为例,采用斯皮尔曼相关性系数对基坑地面沉降的影响因子进行筛选;运用筛选后的8个影响因子建立Stacking深基坑地面沉降预测模型,以验证该方法的适用性。结果表明:Stacking预测模型的平均绝对误差为0.34、平均绝对误差百分比为2.22%,均方根误差为0.13,相较于传统基模型(随机森林、支持向量机和人工神经网络),Stacking预测模型的平均绝对误差、平均绝对误差百分比和均方根误差值皆为最小。In order to improve the prediction ability of machine learning in ground settlement of deep foundation pit, in this study, the authors proposed a ground settlement prediction method based on multi-model combination under Stacking framework. Taking a deep foundation pit in Shenzhen as an example, the Spearman correlation coefficient was used to screen the influencing factors of foundation pit ground settlement, and the eight influencing factors were used to establish the prediction model of ground settlement of deep foundation pit, so as to verify the applicability of this method. The mean absolute error, mean absolute error percentage, and root mean square error of the Stacking prediction model are 0.34, 2.22%, and 0.13, respectively. Compared with conventional base models(random forest, support vector machines, and artificial neural networks),the mean absolute error, mean absolute error percentage and root mean square error values of the Stacking prediction model are minimum.

关 键 词：基坑施工 地表沉降 Stacking模型融合 影响因子筛选 

分 类 号：TU47[建筑科学—结构工程]