基于神经网络的单桩竖向承载力预测-结构工程专业论文.docxVIP

两安建筑科技人学硕Ij学位论文基于神经网络的单桩竖向承载力预测 两安建筑科技人学硕Ij学位论文 基于神经网络的单桩竖向承载力预测 专 业：结构工程 硕士生：李刚 指导教师： 张保印教授 摘 要 本文介绍了神经网络(ANN)的基本概念，概述了多层前馈(BP)网络的基本理 论，重点阐述了多层前馈(BP)网络应用中的若干问题，论证了用神经网络求解桩 基工程问题的可行性，对目前桩基工作特性分析中的一些方法及研究中的主要问 题进行了讨论，对收集到的单桩试桩资料进行了分析。通过对影响桩基承载力因 素的分析并结合BP神经网络的特点，建立了以桩长、桩径、桩入土深度、侧摩 阻加权平均值、桩端承载力等参数为输入层神经元，中间层为l层，中间层的神 经元为2N+1个，输出层为1个(即单桩竖向承载力)的BP神经网络模型。利 用该模型对西安地区钢筋混凝土预制桩、钻孔灌注桩各30根作为样本进行了训 练，并分别对9根钢筋混凝土预制桩和8根钻孔灌注赃进行了仿真预测，将预测 结果与实测值进行了对比分析，其误差都在20％内(大部分在15％V以内)，取 得了较好的结果。 关键词： 神经网络 桩基础 BP网络 竖向承载力 预测 论文类型：应用基础研究 两安建筑科技人学硕l：学位论文Prediction 两安建筑科技人学硕l：学位论文 Prediction of Vertical Bearing Capacity of Single Pile Using Neural Networks Specialty： Structural Engineering Name：Li Gang Supervisor：hof．Zhang Baoyin ABSTRACT this paper has introduced essential concept of Artificial neural network． summarized basal theory of multilayer back—propagation network．put stress on expounding some issue on multilayer back-propagation network in practice， demonstrated the feasibility of resolving pile engineer with Artificial neural network． discussed primary problems about some method and research of characteristic analysis Off pile engineer these days，and analyzed data on test of single pile．It has been established the BP Artificial neural network model that five parameters such as the length of pile，the diameter of pile，the depth of pile embedding soil，weighted average ofside resistance offriction，and bearing capacity ofpile end are import cell， there iS one middle layer,2N+l artificial neural cells of middle layer，and one export layer that is single pile Vertical bearing capacity．Thirty roots of precast reinforced concrete piles and bored and cast．in．place piles in xi’an had been trained with this model，and nine roots of precast reinforced concrete piles and eight roots of cast．in．place piles had been predicted by simulating．When contrasted and analyzed the predicting result and actual numerical value．the error is always less than 20％(mostly within 1 5％)，which has achieved a better result．