【作者】 张研；

【导师】 燕柳斌；

【作者基本信息】 广西大学 ， 结构工程， 2013， 博士

【摘要】 随着经济的飞速增长,地下工程的规模越来越大、施工环境越来越复杂,各种不确定因素越来越多,这就要求与信息化时代相吻合的新的理论与分析方法的提出。鉴于此,本文围绕地下工程岩体非线性行为预测识别与动态智能反馈分析这一主题,将当前流行的高斯过程机器学习方法应用到地下工程当中,提出地下工程非线性系统预测、识别的高斯过程模型,接着与粒子群优化算法相结合,提出粒子群-高斯过程协同优化算法,并将该算法应用到地下工程智能优化反分析,最后在上述研究工作的基础上,综合集成智能优化技术和数值计算方法,融入动态智能反馈分析思想,提出大型地下洞室群动态智能反馈分析方法,为大型地下洞室群的合理开挖支护设计提供科学依据。总体来说,本文主要做了以下几方面研究工作：1.针对传统方法在地下工程岩体非线性系统预测和识别问题上的局限性,研究建立基于高斯过程机器学习技术的地下工程非线性系统预测和识别模型,为实现地下工程非线性系统行为演化过程的准确可靠预测和识别提供一条新的途径。2.由于岩体结构的复杂性,反分析优化问题是典型的复杂、非线性全局优化问题,采用传统的基于梯度信息的解析优化方法只能获得局部最优解,采用随机性全局优化算法时,为了评价随机解的优劣,往往需要借助三维精细仿真对大量的随机解进行成千上万次地适应度评价,难免存在计算耗时过高、计算代价过大的问题。针对上述问题,提出一种基于机器学习与粒子群优化算法的地下工程岩体力学参数智能优化反分析方法,为地下工程岩体力学参数合理获取提供一种新途径。3.开挖后围岩压力的释放情况直接反映了岩体应力卸载程度。在工程实践中,依据新奥法施工理念,当了解围岩压力释放程度时,更有利于发挥围岩的自承载能力,对于判定围岩稳定性和施作支护提供重要依据。为此,将全局寻优性能优异的粒子群算法(PSO)与高斯过程机器学习方法(GP)相融合,结合FLAC3D数值计算软件,提出地下工程岩体力学参数与围岩压力释放率联合反分析的PSO-GP-FLAC3D方法,为岩体力学参数和围岩压力释放率的获取提供一种高效、低成本的方法。4.隧洞施工过程中,监测断面的布置一般滞后于掌子面开挖,监测仪器到位前围岩发生的位移可称为损失位移。损失位移的求取对于围岩稳定性的合理评价具有重要意义,此外,在信息化施工过程中,如果忽视损失位移信息,将给施工安全带来不利影响,并会导致围岩反分析结果出现较大误差。因此,如何获取损失位移成为隧洞围岩稳定性评价和安全施工的关键问题。为此,提出隧洞围岩损失位移求取的PSO-GP-FLAC3D方法,为损失位移的合理获取提供一条新的途径。5.大型地下洞室群动态智能反馈分析的研究对象是复杂的多空间、多维度耦合系统,这种系统存在着数值模型优化、支护方案优化等多种优化问题,这些问题多是全局优化问题。此外,由于岩体结构的复杂性,决策变量与优化指标之间常常存在复杂的非线性隐式关系,借助数值计算技术进行大规模三维精细仿真是解决该问题的常用方式。然而,全局优化问题求解时,大量的数值计算耗时难以令人接受。如何高精度、快速地进行大型地下洞群动态反馈分析,提出合理、实用的动态反馈分析方法,是当前大型地下洞室群施工领域亟待解决的技术难题。另外,现行的动态反馈分析方法多是将本期反演优化后的参数用来预测后期开挖,未将其代入本期开挖进行正算,进而指导对开挖后围岩的局部补强；同时,现有的方法仅对优化后的工程效果进行评价,忽略了优化前信息对工程施工的预测、指导作用。此外,在动态反馈过程中如何对围岩容易失稳的高边墙做出简便、直观地判断,从而为现场施工提供及时的参考,也是当前地下洞室群工程施工过程中亟待解决的问题。针对上述问题,提出一种基于高斯过程的大型地下洞室群动态智能反馈分析方法,为大型地下洞室群安全施工提供一条新途径。6.将上述研究成果应用于某水电站地下厂房大型洞室群动态智能反馈分析,主要针对地下围岩的稳定性分析和支护优化,利用现场监测反馈信息及时优化数值模型和支护方案,实现地下洞室群施工期的反馈分析过程。结果表明：本文提出的大型地下洞室群动态智能反馈分析方法是科学可行的,为大型地下洞室群的科学设计和安全施工提供了参考。更多还原

【Abstract】 With the rapid development of economy, the scale of underground engineering is more and more big, the construction environment of underground engineering is more and more complicated, and the uncertainties of underground engineering is more and more various. It is required that the new theory and analysis methods are proposed to face new challenges. In view of this, rounding the theme that forecasting and identifying nonlinear behavior of undenrground engineering rock mass and dynamic intelligent feedback analysis, the underground engineering nonlinear system prediction model is proposed based on gaussian process (GP) machine learning. Then a novel hybrid optimization algorithm based on particle swarm optimization (PSO) combined GP is proposed. For convenience, the proposed algorithm will be called as PSO-GP that is used for intelligent optimization back analysis. Adopting dynamic intelligent feedback ideas, basing on numerical method combined with intelligent optimization technology, a new method of dynamic intelligent feedback analysis for large underground caverns is proposed. This method puts forward scientific proof for large underground caverns of reasonable excavation supporting design. In sum, the main works and results are listed as follows:1. Aiming to the fact that it is still difficult to reasonably identify and forecast rock mass mechanics nonlinear system behavior in underground engineering, the model based on GP is proposed for identifying and forecasting rock mass mechanics nonlinear system behavior. This provides a new route for accurately and reliably identifying and forecasting rock mass mechanics nonlinear system behavior in underground engineering. 2. Because of the complexity of the rock mass, the back analysis optimization problem is a typical complex, nonlinear global optimization problem. The traditional analytical optimization method based on gradient information only can obtain the local optimal solution. Using stochastic global optimization algorithm, in order to evaluate the quality of random solution, often needs thousands of times fitness evaluation by three-dimensional elaborate simulation, that is a time-consuming and high computational cost problem. Aiming to the problem, the intelligent optimization back analysis method based on GP and PSO is proposed for obtaining rock mechanics parameters in underground engineering. This provides a new route for reasonably obtaining rock mechanics parameters in underground engineering.3. The surrounding rock pressure release situation after excavation directly reflects the degree of rock mass stress unloading. In engineering practice, based on new austrian tunneling method, understanding the degree of surrounding rock pressure release situation can bring into play the surrounding rock bearing capacity and help judge the stability of surrounding rock for applying support. Therefore, a cooperative optimization algorithm based on PSO and GP for back analysis is proposed, then combined the FLAC3D develops a new method called PSO-GP-FLAC3D for joint back analysing of mechanical parameters of rock mass and rock pressure release rate. This provides a new method for high efficiently and low costly obtaining rock mechanics parameters and rock pressure release rate.4. The monitored sections are always assembled behind working face excavation. The displacement induced during this period is called loss displacement that has great significance for reasonably evaluating the stability of surrounding rock. In addition, the construction safety would be affected if you ignore the loss of displacement information, with causing the surrounding rock back analysis results appeared large errors. So, it is a critical problem that how to obtain the loss displacement for tunnel surrounding rock stability assessment and the safety of the construction. Therefore, the method based on PSO-GP-FLAC3D is proposed for obtaining the loss displacemen. This provides a new route for reasonably obtaining the loss of displacemen.5. Dynamic intelligent feedback analysis of large underground caverns is more complicated space and dimensional increase, including numerical model optimization and supporting scheme optimization that is a global optimization problem. The optimization process is a complicated nonlinear problem, with multiple decision variable and optimization indicators, and the relationship between the variables and optimization index is difficult to exactly express. In addition, in order to ensure the accuracy of the calculation, large-scale elaborately three-dimensional simulation is imperative that will cause a huge amount of calculation combining many excavation supporting combination schemes that is a very expensive and time consuming process. Moreover, the current dynamic feedback analysis method predicts the later excavation using the current optimized parameters and not calculates to guide the local reinforcement. At the same time, the existing methods only focus on the optimized effect, ignoring the former information. In addition, how to make simple, intuitive judgment for the high wall in the process of dynamic feedback, so as to provide reference for the construction, is also a problem to be solved in the process of underground cavern group construction. To the above problems, large underground caverns dynamic intelligent feedback analysis method based on GP is proposed that provides an effective route for large underground cavern group safety construction.6. Finally, applying the research fruits mentioned above to the dynamic intelligent feedback analysis of the large underground caverns of certain power station for the stability of surrounding rock in underground analysis and support optimization. Numerical model and support program are timely optimized by monitoring feedback information to achieve feedback analysis process for underground cavern group during construction. The results show that the large underground dynamic intelligent feedback analysis method is feasible that has provided guidance for rational design and safe construction of the large underground caverns.更多还原