基于BP神经网络法对地连墙后土体沉降预测分析——以天津地铁施工为例  

A Settlement Prediction Analysis of Soil Behind Diaphragm Retaining Walls based on BP Neural Networks——Take Construction of Tianjin Metro as an Example

作　　者：刘戈[1] 吴立新[1] 

机构地区：[1]天津城建大学土木工程学院,天津300384

出　　处：《沈阳建筑大学学报（自然科学版）》2013年第5期834-840,共7页Journal of Shenyang Jianzhu University：Natural Science

基　　金：国家自然科学基金项目(40972170);天津市建委重点基金项目(2010-11)

摘　　要：目的研究深基坑地连墙墙后土体沉降的关联因素以及预测深基坑观测点的沉降数值,解决监测中的空白现象.方法以天津地铁5号线张兴庄站为例,应用基本负梯度下降数学理论,并基于实际工程特点和MATLAB程序平台,对收集到的数据进行前期处理,经过对影响地连墙墙后土地沉降的关联因素的分析,编制BP神经网络预测模型程序,从而达到准确预测沉降数据的效果.结果 BP神经网络预测模型的训练结果和实际监测值高度吻合,在预测结果中,被破坏的观测点的预测结果与实际结果的误差小于1%.结论在深基坑施工过程中,利用主成分系统分析得出关联因素,基于BP神经网络预测模型,可以利用正常使用的监测点对已破坏的监测点进行模拟预测,在有效时间内相对准确地预测出沉降数据,为施工过程提供科学依据.In the process of construction, due to changes with the weather, large construction vehicles com- pacted ,construction materials piled and others, some observation points of soil settlement are destroyed, which would lead to the monitoring data not to be comprehensive, refer to the study of associated factors of settlement of soil wall of deep foundation pit and numerical prediction of settlement observation point is the effective way to solve the monitoring blank phenomenon. Based on actual project characteristics and MAT- LAB program platform, this article takes Tianjin Metro Line 5 Zhang Xingzhuang station as an example and applies basic mathematical theory of the negative gradient descent to analyze the associated factors which af- fect the soil behind the wall and program BP neural network prediction model, which is to achieve accurate prediction of the effects of settlement data. In this article, BP neural network prediction model results and the actual monitoring data are highly consistent. In predicting results, the error of prediction and actual monito- ring data is less than 1%. In the process of construction, using associated factors which is obtained by princi-pal component analysis, and based on BP neural network prediction model, this article refers to the normal monitoring points for the forecast data, which have been destroyed. In the effective time, this article accurate- ly predicts settlement data and provides a scientific basis for the construction process.

关 键 词：预测分析 神经网络 关联因素 土体沉降 

分 类 号：TU476.3[建筑科学—结构工程]