【作者】 吴晓恩；

【导师】 隆威；

【作者基本信息】 中南大学 ， 地质工程， 2008， 硕士

【摘要】 我国软土分布十分广泛,它的主要特征是:天然含水率高、天然孔隙比和压缩性大等。软土地基具有承载能力低、沉降量大、固结完成时间长等不利的工程特性,在其上建造构筑物广泛存在的软土地基沉降问题一直以来都是一个技术难题。因此深入探讨软土地基的沉降发展规律,利用有限的沉降实测数据,选取合理的预测模型及方法预测地基的后期沉降(包括最终沉降),对控制施工进度,指导后期的施工组织与安排,具有重要的理论与工程实际意义。本文介绍和探讨了塑料板排水堆载预压法在深圳盐田港西港区纳泥塘的软基加固工程中的应用、设计方案、现场观测成果,并对此方法加固效果进行了检验和分析,完全满足设计要求。详细阐述了人工神经网络的原理和实现过程,针对实际工程的不同条件和要求,本文采取了两种不同的神经网络建模法。一种是把各影响因素同沉降的关系用神经网络隐式表达,通过建立反映软基沉降影响因素与软基沉降量的映射关系的BP向前型网络模型,用已有的沉降观测资料对网络进行学习训练。预测时,由已知外界影响因素推断此时的沉降;另一种方法不考虑沉降的各影响因素,而是建立反映当前沉降同过去各沉降历史值间关系的Elman反馈型网络模型,通过高度非线性的曲线拟合,推求工程后期的沉降量。通过样本训练验证以及与传统的分层总和计算方法和曲线拟合法中的双曲线法、指数曲线法、泊松曲线法、Asaoka法的对比,对拟合预测结果进行检验,使其具有统一的评价标准。结果表明,在预测深圳盐田港西港区纳泥塘地区软基沉降方面,BP神经网络模型和双曲线法的预测效果最好。由于缺少最终沉降量实测值,Asaoka法预测效果还需在该地区作进一步验证。更多还原

【Abstract】 The distribution of soft soil is very extensive in China,the characteristics of it are:high natural moisture content ratio,high natural void level and high compression ratio.The unfavorable characters of soft soil foundation are:the low carrying capacity,high settlement and long consolidation time.The soft soil consolidation problem of the buildings are set up on it which construct on the soft soil is a technical tickler.So it is very important to theory and actuality by probing the regularity of soft soil,forecast the final consolidation by the forecast model and methodology from the actual settlement data.A study is carried out to the application scope of stak preloading and dynamic compaction in soft foundation treatment of Nanitang in Yantan port of west port area of Shenzhen,the design of construction parameter,local observation.The examination and analysis are made on the reinforcement effect of this method.The result shows that the analyzer fulfills all design requirements,expounds the principle and implementation procedure of ANN.Two different modeling methods of neural network are used in the paper for different conditions and requirements in practical engineering.One is BP using implicit formula to express the connection between factors and settlement.The current settlement can be concluded by known factors when predicting.The other is Elman not considering the factors of settlement,but establishing the neural network model for the connection between current settlement and the past.Then the late settlement can be predicted by highly nonlinear curvefitting.Verify and contrast to the traditional layerwise summation method,some methods of curve fitting method such as hyperbolic method,exponential method,poission curve method and Asaoka method,and the sample training, the check are used to discuss the results of fitting and prediction so that all these methods can be evaluated by using an unified standard. It is shown that BP and the hyperbolic method are considered as the most effective ones among methods on settlement prediction in soft foundation treatment of Nanitang in Yantan port of west port area of Shenzhen.Because scarcity of the test final settlements date, The prediction effection of Asaoka method should be further validated in more projects in this area.更多还原