基坑地墙渗漏指标预测及风险等级评价方法  

Prediction of Leakage Index and Evaluation of Leakage Risk for Foundation Pit Wall

作　　者：张迪[1] 孙峰[1] 叶晓剑 曹刚 邱子琪 曹子君[4] 洪义[3] 王立忠[3] 王立林[3] ZHANG Di;SUN Feng;YE Xiaojian;CAO Gang;QIU Ziqi;CAO Zijun;HONG Yi;WANG Lizhong;WANG Lilin(China Railway Siyuan Survey and Design Group Co Ltd,Wuhan 430063,China;Fuyang City Construction Investment Group Co Ltd,Hangzhou 311400,China;Hainan Institute,Zhejiang University,Haikou 572025,China;Institute of Smart City and Intelligent Transportation(Institute of Urban Rail Transportation),Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]中铁第四勘察设计院集团有限公司,湖北武汉430063 [2]杭州富阳城市建设投资集团有限公司,浙江杭州311400 [3]浙江大学海南研究院,海南海口572025 [4]西南交通大学智慧城市与交通学院(城市轨道交通学院),四川成都611756

出　　处：《土木工程与管理学报》2025年第1期20-26,共7页Journal of Civil Engineering and Management

基　　金：国家自然科学基金(52238008,52122906)。

摘　　要：太阳能、风能等清洁能源需要占用有限的地表空间,而人类对地下空间的开发利用间接促进清洁能源的发展。本文选取基坑事故中的地墙渗漏问题,运用有限元软件ABAQUS模拟1000组不同程度的基坑渗漏环境,并提取观测井孔隙水压力、地连墙侧向位移、地面沉降三种物理量在地墙渗漏环境下7 d的变化数据。数据经过主成分分析(PCA)降维后,被用于构建人工神经网络(ANN)、循环神经网络(RNN)、长短期记忆网络(LSTM)、门控神经网络(GRU)四类神经网络模型,实现基坑渗漏的关键变量的预测,以及关键变量的风险等级评估。本文还尝试用决策树与随机森林预测基坑渗漏风险等级。对比回归与分类两种风险等级评估思路,分类方法可以快速实现渗漏等级的预测,回归方法可以实现渗漏关键变量的预测。Clean energy sources such as solar and wind power require the use of limited surface space,while human’s development and utilization of underground space indirectly promote the growth of clean energy.This paper focuses on the issue of foundation pit wall leakage in excavation accidents.Using the finite element software ABAQUS,1000 different leakage scenarios were simulated,and data on three physical quantities—pore water pressure in observation wells,lateral displacement of foundation pit walls,and ground settlement were extracted over a 7-day period in these environments.After dimensionality reduction using principal component analysis(PCA),the data were used to construct four types of neural network models:artificial neural network(ANN),recurrent neural network(RNN),long short-rerm memory(LSTM),and gated recurrent unit(GRU).These models were employed to predict key variables of excavation leakage and assess the risk levels of these variables.The paper also explores the use of decision tree and random forest for predicting leakage risk levels.Comparing the regression and classification approaches for risk assessment,the classification method quickly predicts the leakage levels,while the regression method predicts the key variables of leakage.

关 键 词：基坑渗漏 安全预警 机器学习 神经网络 主成分分析 

分 类 号：U231.4[交通运输工程—道路与铁道工程]