【作者】 徐冲；

【导师】 刘保国；

【作者基本信息】 北京交通大学 ， 岩土工程， 2011， 博士

【摘要】 分岔隧道是一类由大跨变断面段、连拱段和小间距段共同构成的组合隧道形式,其线形设计上的多断面、多尺度和施工工序及围岩荷载频繁转换的特点导致其开挖支护体系在复杂应力空间中表现为一个难度自增值系统。如何对该系统进行全局性、准确性的描述和决策,并提出一套能够服务于此类隧道开挖与支护方案优化的普适性方法是本文需要解决的核心问题。在此,本文以胶州湾海底隧道陆域段左线分岔隧道为工程背景,综合三维数值计算、仿生优化、核机器学习和突变理论等方法,提出分岔隧道洞室稳定性分析与全局智能优化的新方法。该方法首先将仿生优化算法与弹塑性数值计算相融合,完成工程区域围岩力学参数反演；其次,基于主元分析和位移增量基础信息分别建立多权重综合评价体系和洞室稳定性的突变判据,继而采用新的智能全局优化方法在开挖与支护方案各种组合的全局空间下搜索出最优施工方案,成功地解决了分岔隧道开挖与支护方案优化过程中的多目标决策和洞室稳定性评价问题。具体研究内容及成果如下：1.核学习机方法及仿生优化算法研究(1)作为第三代核机器代表的高斯过程具有小样本、概率推理、泛化性能好的优势,将其应用于边坡和隧道位移非线性时间序列的建模分析表明：高斯过程较人工神经元网络和支持向量机在网络训练精度和预测精度上都有显著提高。(2)利用模拟退火算法兼具“上山性”和“下山性”的特性克服标准粒子群算法易陷入局部最优的缺陷,提出基于二者串行编程的混合优化算法—粒子群-模拟退火算法。其函数测试结果表明：该算法能够适用于高非线性问题的优化求解。(3)为克服传统共轭梯度法在核参数优化过程中的不足,实现混合算法与高斯过程的程序耦合。针对典型滑坡变形预测问题,与基于遗传算法进化的高斯过程等算法进行了对比研究,前者在程序简易性和测试效果两方面均有较大优势。2.基于计算智能方法的分岔段开挖与支护方案优化研究(1)采用FLAC内置FISH语言,构建了粒子群算法与FLAC2D耦合的进化有限差分方法—粒子群-FLAC2D,继而提出了围岩物理力学参数的位移智能反演方法及实现步骤。工程实践表明：该方法在计算效率和优化精度上都具有较大优势。(2)利用主元分析方法解决多目标优化过程中优化指标的权重分配问题,在保存评价指标基础信息的前提下既可达到降维、简化评价体系的目的,又可以避免权重确定的人为因素；同时,首次提出将基于位移增量步的突变判据作为洞室稳定性评价新标准和多目标决策优化过程的约束条件。(3)针对洞室开挖与支护全局优化问题,为避免大量三维数值计算带来的成本消耗,提出了开挖与支护方案的全局智能优化方法：粒子群-模拟退火-高斯过程-数值计算方法。该方法融合了计算智能全局并行搜索和学习机黑箱推理的特性,能高效率、高精度的处理多目标决策问题。同时,在工程方案决策过程中避免了于几种预选方案中择优的不合理性和局限性,为解决此类工程决策问题提供一种新的有效途径。3.变速段的线形优化及断面轮廓设计研究首先结合基于位移增量信息的突变判据和传统位移、应力指标分析,给出变速段线路过渡方案应以台阶式为最佳过渡方式的结论；其次,考虑匝道及主隧的不同行车速度、发动机制动加速度等因素,提出了线路汇流和分流情况下修正了的变速段长度计算公式和台阶过渡方式下的线路平面线形设计公式；最后,基于作图法原理,给出了变速段各台阶断面的空间布局设计方法及其轮廓设计公式。4.胶州湾海底隧道陆域分岔隧道设计施工优化研究基于进化有限差分方法及其相应的智能位移反演步骤,成功地实现了工程区域围岩力学参数的位移智能反演；接着,集成全局智能优化方法(程序)和ANSYS、TECPLOT等常用优秀前后处理软件技术完成了分岔段的开挖与支护方案优化和洞室稳定性分析,在施工技术许可的前提下提出了合理的支护设计参数和施工方法,继而根据线形优化设计方法确定出变速段的断面尺度及其空间布局。更多还原

【Abstract】 Forked tunnel is a kind of combining from tunnel that consists of large-arch, multi-arch, neighborhood and transition tunnel. The special structure type and conversion of stress and operation sequence frequently leads to the fact that its excavation and support system is an accumulative difficulty system in complex stress space in tunnel construction. This paper concentrates on how to bring up a global and accuracy description and assessment on the design and construction scheme and further get a universal method. For this purpose, under the background of construction of kiaochow bay subsea tunnel, stability analysis and intelligent global optimization method for forked tunnel is proposed by integrating different theory and technology, such as 3D numerical simulation, bionic optimization, kernel machine learning and catastrophe theory. Specific steps are as follows:Bionic optimization algorithm firstly couples with FLAC for displacement back-analysis of mechanical parameters of surrounding rock. Moreover, PCA and catastrophe criterion are used to cope with weight distribution and stability evaluation respectively. Finally, the global optimum excavation sequence and support parameters are obtained by the new intelligent global optimization method. Thus, the problem mentioned above could be solved successfully based on the new method. Detailed research results include the followings:1. Study on the kernel machine learning and bionic optimization algorithm(1) Gaussian process (GP), the third generation kernel machine, has advantages of small samples, probabilistic reasoning and good generalization capability. The analysis results of displacement nonlinear time series of tunnel and slope indicate it notably improves on the accuracy and efficiency of prediction compared with ANN and SVM.(2) Simulated annealing (SA) with characteristics of’uphill’and’downhill’ simultaneously is introduced to improve the standard particle swarm optimization (PSO) for modifying the population diversity and then the new coupling algorithm PSO-SA is proposed. Its application results of function test state clearly that the algorithm is applicable to solve the highly nonlinear problems.(3) For overcoming the shortcomings of conjugate gradient during the process of optimizing the kernel function of machine learning, the PSO-SA instead couples with GP for forming the new PSO-SA-GP algorithm. Compared with the GA-GP algorithm, the new algorithm demonstrates advantages in two areas, namely simplicity and efficiency of code, aiming at predicting the displacement of typical landslide.2. Study on excavation and support optimization based on computation intelligence(1) PSO is used to couple with FLAC2D in order to form the new evolutional finite difference method. Then it is adopted to cope with the back-analysis of mechanical parameters of surrounding rock with the advantages of computational efficiency and optimization accuracy.(2) PCA and catastrophe criterion are used to handle the weight distribution and stability evaluation during the process of optimization on excavation and support scheme respectively. The former can achieve the goal of simplifying the evaluation system and dimension reduction and the latter can be served as a new evaluation index and constraints for multi-objective decision making.(3) Aiming at the global optimization of excavation and support scheme, a new PSO-SA-GPR-FLAC3D method with global parallel search and black-box reasoning is put forward for reducing the cost of large-scale 3D numerical simulation. It can handle the multi-objective decision making well and used as a new way for similar engineering.3. Study on line-shape optimization of transition section and contour design of cross-sectionFirstly, the stepped layout is decided as optimum transition mode based on catastrophe criterion and traditional analysis indexes. Secondly, considering the different speed between main tunnel and ramp tunnel new modified formulas for the lengths of transition section and horizontal line-shape are proposed. Finally, the spatial layout and contour design method of stepped cross-section is got by mapping principle.4. Study on the optimization of design and construction of forked section of kiaochow bay subsea tunnelMechanical parameters of surrounding rock in tunnel site are obtained by PSO-FLAC2D. Then the optimization of excavation and support scheme and stability of tunnels are done by the integrated global intelligent method including the AI, ANSYS and TECPLOT etc. Finally, reasonable supporting parameters, construction sequence, section scale and its spatial layout are provided for builders.更多还原