基于动态神经网络NARX时间序列的双排桩基坑变形预测  

Deformation prediction of double-row pile foundation pit based on dynamic neural network NARX time series

作　　者：侯福昌 曾家俊 江杰[3,4] 李结全 范懿文 HOU Fuchang;ZENG Jiajun;JIANG Jie;LI Jiequan;FAN Yiwen(Guangxi Ruiyu Construction Technology Co.,Ltd.,Nanning 530004,China;School of Civil Engineering and Architectural,Guilin University of Technology,Guilin 541004,China;School of Civil Engineering and Architecture,Guangxi University,Nanning 530004,China;Key Laboratory of Disaster Prevention and Structural Safety of Ministry of Education,Nanning 530004,China;School of Civil Engineering Guangxi Polytechnic of Constrnction,Guangxi Polytechnic of construction,Nanning 530004,China)

机构地区：[1]广西瑞宇建筑科技有限公司,广西南宁530004 [2]桂林理工大学土木与建筑工程学院,广西桂林541004 [3]广西大学土木建筑工程学院,广西南宁530004 [4]工程防灾与结构安全教育部重点实验室,广西南宁530004 [5]广西建设职业技术学院土木工程学院,广西南宁530004

出　　处：《广西大学学报（自然科学版）》2024年第1期49-59,共11页Journal of Guangxi University（Natural Science Edition）

基　　金：国家自然科学基金项目(52068004);广西自然科学基金项目(2018GXNSFAA050063);广西交通运输行业重点科技清单项目(JZY2020KZD02)。

摘　　要：针对目前基于含基本假设或经验公式的传统土力学计算方法,不能有效地反映具有多因素交叉性以及时空性的基坑变形规律,而监测数据时间序列能够真实地表现基坑土体变形的演变,以南宁市亭洪路72号河南水厂住宅小区危旧房改造项目双排桩基坑工程为依托,考虑开挖深度和土体暴露时间这2个因素对监测时间序列的影响,提出一种带有外部输入的非线性自回归(NARX)动态神经网络时间序列模型,多方位预测关键断面重要测点的竖向位移和水平位移。结果表明:预测值和实际监测数据的变化趋势具有较好的一致性,且竖向位移预测值与实际监测值的预测残差小于1.0 mm,水平位移预测残差小于0.3 mm。该模型预测效果良好,同时验证了此模型应用于双排桩基坑变形动态分析的可行性。In view that the traditional soil mechanics calculation method based on basic assumptions or empirical formulas usually cannot effectively reflect the deformation law of the foundation pit with multi-factor intersection and spatial and temporal characteristics,and then the time series of the monitoring data can truly express the evolution of the deformation of the foundation pit soil body,in this paper,based on the double-row pile foundation pit project of residential distric renovation of dilapidated buidings of Henan Water Plant,which is located in No.72 Tinghong Road,Nanning,Guangxi,considering the influence of two factors,excavation progress and soil exposure time on the monitoring time series,we propose a nonlinear-auto-regressive model with exogenous inputs(NARX)time series dynamic neural network model,which predicts the vertical and horizontal displacements of the key points in key sections in multiple directions.The results show that the trend of the predicted values and the actual monitoring data have good consistency and the residual difference between the predicted one and the practical one is less than 1 mm.Moreover,the residual difference between the horizontal displacement and the predicted value is less than 0.3 mm.The model has good prediction effect,and it also verifies the feasibility of this model applied in the dynamic analysis of the deformation of the double-row pile pit.

关 键 词：动态神经网络 时间序列 预测模型 双排桩 基坑变形 

分 类 号：TU473.1[建筑科学—结构工程]