基于遗传神经网络算法的基坑土体参数反演和基坑沉降预测研究

发布时间：2018-03-24 01:09

本文选题：数值模拟　切入点：均匀设计　出处：《山东大学》2017年硕士论文

【摘要】：目前,我国已经成为世界上兴建地铁线路和车站最多的国家,车站基坑工程引起周边沉降的机理非常复杂。随着计算机技术和数值计算理论的发展,数值分析方法开始逐渐应用到基坑开挖分析中,然而数值计算结果较多地依赖于模型参数的选取,忽略了基坑工程的特定条件,模型参数将很难准确确定。随着现代智能算法的兴起,智能算法开始运用到基坑开挖中。遗传神经网络算法兼具非线性拟合和非线性寻优的特点,遗传神经网络算法对参数进行反演分析具有很强的实用性。本文依托青岛二号线燕儿岛车站基坑开挖工程,通过实测分析、数值模拟、参数反演、沉降预测相结合的研究方法,对地铁车站在开挖过程中周边测点的沉降变形规律进行了研究,并且对土体的强度参数进行了反演分析。本文的主要研究内容和成果如下:(1)对车站基坑周边的地下水位、围护桩侧移、坑外地表沉降的实测数据进行分析,提出了这些监测项目随基坑开挖过程变化的原因,确认了数值模拟方案中不考虑降水措施。(2)设计了基坑的围护桩、混凝土支撑、钢支撑、锚索、围护桩与土体接触面等的梳子模拟方案,基于FLAC3D软件编制了基坑开挖的计算程序,对不能确定的弹性模量,根据经验对土体弹性模量取值,计算了在该工况下,坑外测点随着基坑开挖的沉降值,与实测结果对比,验证了所建立的模型模拟方案的合理性。(3)为了合理设计试验样本、降低试验次数,引入均匀设计法,设计了学习样本;利用遗传神经网络算法的优点,提出基于遗传神经网络算法的多参数反演分析思路。基于MATLAB软件,编制了遗传神经网络算法参数反演程序。最后通过某基坑工程实例,反演得到了该工程较为合理的土体弹性模量,验证了程序和分析思路的可靠性和准确性。(4)提出沉降的动态预测实施步骤,基于基坑外测点的实测数据和遗传神经网络算法,编制动态预测程序,训练得到了前期沉降与后期沉降的非线性映射关系,并结合某一测点,对其沉降值进行预测。结果表明,基于遗传神经网络算法的动态沉降预测方法具有很高的可靠性。
[Abstract]:At present, China has become the country with the largest number of subway lines and stations in the world, and the mechanism of surrounding settlement caused by station foundation pit engineering is very complicated. With the development of computer technology and numerical calculation theory, The numerical analysis method is gradually applied to the excavation analysis of foundation pit. However, the numerical calculation results depend on the selection of model parameters and ignore the specific conditions of foundation pit engineering. With the rise of modern intelligent algorithm, intelligent algorithm has been applied to excavation of foundation pit. Genetic neural network algorithm has the characteristics of nonlinear fitting and nonlinear optimization. The genetic neural network algorithm has strong practicability for parameter inversion analysis. Based on the excavation project of Yanerdao station in Qingdao No. 2 line, this paper combines the research method of actual measurement analysis, numerical simulation, parameter inversion and settlement prediction. In this paper, the law of settlement and deformation of the site around the subway station during excavation is studied, and the strength parameters of the soil are inversely analyzed. The main contents and results of this paper are as follows: 1) the groundwater level around the foundation pit of the station. Based on the analysis of the measured data of the lateral displacement of the retaining pile and the surface settlement of the pit field, the reasons for the variation of these monitoring items with the excavation process of the foundation pit are put forward, and it is confirmed that the retaining pile of the foundation pit is designed without considering the dewatering measure .Y2) in the numerical simulation scheme. The comb simulation scheme of concrete bracing, steel bracing, anchor cable, retaining pile and soil contact surface, etc. Based on FLAC3D software, the calculation program of foundation pit excavation is compiled, and the value of soil elastic modulus is obtained according to experience for the uncertain elastic modulus. Under this condition, the settlement value of the measuring point outside the pit with the excavation of foundation pit is calculated. Compared with the measured results, the rationality of the model simulation scheme is verified. In order to design the test sample reasonably and reduce the test times, the uniform design method is introduced. Based on the advantages of genetic neural network algorithm, the idea of multi-parameter inversion analysis based on genetic neural network algorithm is put forward, which is based on MATLAB software. The parameter inversion program of genetic neural network algorithm is worked out. Finally, through an example of foundation pit engineering, the reasonable elastic modulus of soil is obtained. The reliability and accuracy of the program and the analysis idea are verified. (4) the dynamic prediction procedure of settlement is put forward. Based on the measured data of the site outside the foundation pit and the genetic neural network algorithm, the dynamic prediction program is compiled. The nonlinear mapping relationship between pre-settlement and post-settlement is obtained, and the settlement value is predicted by combining a certain measuring point. The results show that the dynamic settlement prediction method based on genetic neural network algorithm is highly reliable.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U231.3;TU433

【相似文献】