【作者】 陈斌；

【导师】 卓家寿； 刘宁；

【作者基本信息】 河海大学 ， 工程力学， 2001， 博士

【摘要】 本文针对岩土工程反分析中使用的传统贝叶斯法中存在的缺陷，提出了扩展贝叶斯法，从概率论和数理统计的原理出发，建立了基于决策信息论中AIC准则和最大熵准则的岩土工程随机反演的准则函数，解决了传统贝叶斯法的观测信息与先验信息的匹配问题，并从形式上完成了目前常用的最小二乘法、最大似然法及传统贝叶斯法准则函数的统一，并可由观测数据的质量和数量进行预测模型的辨识。 论文应用最大熵原理推导了最大似然法和传统贝叶斯法的准则函数，从信息论的角度为这两种反演分析方法的发展开拓了更广阔的应用空间。 论文以能较好模拟土石坝非线性力学行为的Duncan-Chang E-B模型为基础，通过引入应力空间下的自然坐标系，分解应力球量与应力偏量，从而导出一个在有限元计算中，位移对参数偏导数的闭合形式公式。并提出了在参数敏感性分析中应考虑位移值的影响，建议采用位移加权的方法来解决这一问题。 论文采用面向对象的软件PowerBuilder作为开发平台，运用动态链接库的原理，建立了一套面向对象的反演分析法，定义了对象类之间的关系，确立了有限元的消息流，使程序的界面更加友好，程序操作更为直观。 所建立的计算模型和分析方法，通过了数值算例的验证，并成功地应用于小浪底水利枢纽工程安全监控系统中。从而也表明本文提出的模型和分析方法基础可靠、仿真性强，具有广阔的应用前景。更多还原

【Abstract】 ABSTRACTAccording to the defect of the conventional Bayes’ method in the geomechanical engineering back analysis, expanding Bayes’ method was established in this paper. With the applications of the probability theory and the mathematical statistics principle, the geomechanical engineering random back analysis objects function was founded based on the AIC criterion of the decision information theory and the maximum entropy criterion, a the matching question between the observed information and the previous information of the conventional Bayes’ method were proposed. In addition, the least square method, maximum likelihood method and conventional Bayes’ method were unified in form. The predicted model can be identified by quality and quantity from the observed values.Two objects functions were deduced, which of the maximum likelihood method and the conventional Bayes’ method, with the application of principle of maximum entropy. Under the information theory, more applications are provided to those back analysis methods.Based on the Duncan-Chang E-B model, which can simulate earth-rock dam’s non-linear mechanics behavior well, the paper introduced the nature coordinate system to the stress space. In this way, hydrostatic stress and stress deviator were decomposed each other. Therefore, a partial derivative formulation about displacement and parameter in finite element evaluation was derived in a closed form. On the same time, a displacement-weighted method was suggested to resolve the behavior of displacement factor in the parameter sensitivity analysis.Using the soft of object-orient, PowerBuilder, and DLL theory, an object-oriented back analysis method was studied, which defined the relationship of object classes, established a message flow of finite element, and also made program of friendly interface and efficient operation.The models and analytical methods presented above were verified by some numerical examples, and were applied successfully to the Huanghe XiaoLangDiHydro-Junction Project softy monitoring system. The results show that the models and methods proposed in this paper have so good foundation and emulation that a wide range of application can be provided.更多还原