【作者】 黄传胜；

【导师】 张家生；

【作者基本信息】 中南大学 ， 道路与铁道工程， 2011， 博士

【摘要】 深基坑工程在施工期会产生基坑围护结构的位移、坑底隆起、周边地表沉降等变形,如变形值过大会引起基坑失稳或周边建(构)筑物开裂等工程事故。影响深基坑变形的因素很多,属于典型的非线性问题。按照信息化施工的需要,深基坑工程施工过程中必须开展施工监测,并根据监测资料对基坑的变形进行预测。本文以广州地铁三号线燕塘站深基坑工程(迄今为止广州地区开挖深度最大的基坑工程)为背景,采用有限元法对深基坑开挖施工引起的变形进行了分析,建立了基于灰色系统、马尔科夫链及人工神经网络的变形预测模型,对各种预测模型进行了分析,通过实际验证,指出了各模型的适用性。本文所做的主要工作如下：(1)分析了燕塘站深基坑工程施工的特点和难度,制订了明挖部分深基坑工程施工方案和监测方案。确定采用地下连续墙+内支撑的支护方案。对施工期基坑变形进行了全程连续监测。(2)对地铁燕塘站深基坑开挖进行了三维数值模拟分析,并深入分析了模型的尺寸效应和空间效应,确定了模型最终尺寸。桩墙体水平位移、支撑轴力以及基坑周围地表沉降等分析计算结果与实测数据吻合较好；分析了施工工序及各道工序对基坑位移的影响；通过对比分析,发现采用硬化土模型、设置桩-土界面接触单元所得计算结果更加符合工程实际。(3)验证了新陈代谢模型的建模适用性,建立了用于短期预测的新陈代谢GM(1,1)模型和灰色马尔科夫链模型,对基坑变形值进行了短期预测。结果表明,短期预测新陈代谢GM(1,1)模型和灰色马尔科夫链模型的预测精度都可以满足施工期预测要求,可优先采用灰色马尔科夫链模型。(4)建立了适合于中长期预测的新陈代谢残差GM(1,1)预测方法、灰色马尔科夫链残差预测方法,并利用这些方法对基坑变形值进行了中长期发展预测。预测结果表明,这两个模型的预测精度均可以满足施工期预测要求,但灰色马尔科夫链残差模型更合适。(5)分析了影响深基坑变形的各影响因素,选择土体内摩擦角、土体粘聚力、土的重度、地下水位、渗透系数、深基坑开挖深度、内支撑层数等7个因素为深基坑变形BP网络预测模型的影响因素。分别探讨了基于时间序列和基于各影响因素的深基坑施工期变形BP网络的建模方法,利用所建模型进行沉降变形预测,模型预测精度都较好。(6)将灰色马尔科夫预测模型、灰色马尔科夫残差模型、基于时间序列BP神经网络模型、基于各影响因素BP神经网络模型进行对比分析,各预测模型与现场监测值均吻合的较好,都能应用于施工期深基坑变形动态预测,其中基于各影响因素的BP神经网络模型预测值与现场监测值最接近。BP网络模型的预测精度比灰色马尔科夫模型稍高。更多还原

【Abstract】 Deep excavation engineering construction will produce pit supporting structure displacement, foundation-pit bottom uplift, the surrounding ground settlement and etc, big deformation will cause instability of foundation pit or crack of the surrounding structures and other engineering accidents. There are many factors influence the deep foundation pit deformation, which belong to the typical nonlinear problems. According to the need of informatization construction, construction monitoring must be carried out in the process of deep foundation pit engineering construction, and deformation forecast must be made according to the monitoring datas of foundation pit. This paper is taking deep foundation pit engineering of Yan Tang station of No.3 Guangzhou metro line for example (so far the largest deep excavation engineering in guangzhou area). Finite element method is used to analyze the deformation caused by deep foundation pit excavation construction, a predicting model is constructed based on grey system, markov chain and artificial neural network to analyze various deformation forecast model. The applicability of various model is made through actual test and verify check. The main works in this paper are as follows:(1) The properties and difficulties of the deep excavation construction of Yantang station were analyzed. The construction scheme and the monitor scheme of open-cut part of deep foundation pit of Yantang subway station were made. The supporting scheme underground continuous wall combined with inner supporting are adopted. The deformation of the deep foundation pit is monitored continuously during the whole construction.(2) Three dimensional mathematical analyze was carried out on the excavation of the deep foundation pit of Yantang subway station. The model’s size effect and space effect were deeply studied and the final dimensions of the model were confirmed. The analysis result of piles and walls’horizontal displacement, supporting axis strength and the surrounding earth surface deformation fit well with in-situ survey value. The effects of construction working procedure and each working step to deformation of the deep excavation were analyzed. According to the contrast analysis, the calculation results fit well with actual project results by using rigidification soil model and setting piles-soil interface contact element to the calculation model.(3) The modeling applicability of the metabolism model was verified. Metabolism GM (1,1) model and gray markov chain models for short-term forecasting were established, short-term prediction of the deformation of foundation pit was done. The results show that predict accuracy from which the short-term forecast metabolism GM (1.1) model and gray markov chain models predict can meet the requirements which the construction need and grey markov chain model can be firstly considered.(4) For medium and long-term forecast, metabolism residual GM (1,1) forecasting methods and gray markov chain residual forecasting method were established. The deformation forecast values for medium and long term development was predict by using these methods. The predicted results show that the prediction accuracy of the two models both can satisfy the demand of the construction need, but gray markov chain residual model was more appropriate.(5) Factors which affect the deformation of deep foundation pit were analyzed.7 factors was chosen to be the effect factors for BP neural network predict model. They are the soil internal friction angle, soil cohesion force, unit weight of soil, groundwater table, the permeability coefficient, excavation depth of deep foundation pit, number of interior support layers. The BP network construction modeling methods based on time sequence and based on each influence factor are discussed respectively. The deformation prediction was made by using the models, the prediction precision of the models are good.(6) Comparison analysis among the grey markov forecast model, gray markov residual model, BP neural network model based on time sequence, BP neural network model based on the influence factors were carried out. The values from analysis forecasting model are fit well with the in-situ monitor values and all the models can be applied for deformation dynamic prediction of deep foundation pit during construction. The calculating values from the BP neural network model based on each effect factor most closely meet the field monitoring values. The prediction accuracy of BP network model is slightly higher than gray markov model.更多还原