堆石坝参数反演的蚁群聚类RBF网络模型

 将蚁群算法与径向基(RBF)网络相结合,提出一种用于堆石坝力学参数反演的蚁群聚类径向基网络模型。该模型用蚁群聚类算法搜索RBF网络基函数中心,模拟蚁群觅食聚类的概率转移特性,所得到的聚类结果类间离散度和比传统K均值聚类结果小,能够得到更合理的基函数中心,从而获得较准确的坝体参数和位移之间的非线性映射关系。在进行参数灵敏度分析的基础,对一座堆石坝的反演分析表明,蚁群聚类RBF网络模型可有效地求解堆石坝多参数反演问题,反演结果优于BP网络模型和K均值RBF网络模型。     

岩土工程反分析的计算智能研究

 传统岩土工程学科的研究方法及思路限制了岩土工程反分析技术,乃至整个学科的发展,因此,从研究思维变革的根本上着手,把现代智能科学中的新发展分支—计算智能引入岩土工程反分析,可望使其产生一个飞跃性突破,以至带动整个岩土工程学科的发展。基于此思路,本文把计算智能方法引入反分析领域(本文反分析指位移反分析),全面而系统地进行了反分析的计算智能研究。 岩土物性参数反演可分为正反分析(优化反分析)及逆反分析两种。针对正反分析目标函数的复杂性,把进化计算引入正反分析研究,提出了进货反分析方法。基于前人的研究及对算法机理的分析,提出了一种快速收敛遗传算法反分析方法。首次结合进化规划、人工免疫系统原理及有限元计算提出了一种新的免疫进化规划反分析方法,标准算例证明它是一种相当理想的正反分析方法。从系统逆辨识的全新角度观察反分析研究,提出了基于神经网络的逆反分析思路。并提出了一种结构及权值同时进化的新型进化神经网络模型,并由它进行了进化神经网络逆反分析研究。 为了进行模型辨识,基于岩土本构模型的统一表达式,提出了一种新的岩土本构模型类内辨识的进化方法。基于分子生物学及现代进化论观点,创造性地提出了一种既模拟生物种间竞争又模拟生物种内竞争的生态竞争进化模型。基于此,提出了一种岩土本构模型类间识别并同时反演其物性参数的新算法。 根据位移时序分解原理,以灰色系统拟合位移序列的趋势,以本文提出的网络结构及算法参数同时进化的进化神经网络逼近序列的趋势偏离量,首次提出了基于灰色系统—进化神经网络的位移时序混合智能建模新方法。并通过两个工程实例验证了该混合智能方法的良好位移建模及预测能力。 最后,采用天生桥一级电站工程、某抽水蓄能电站工程及三峡永久船闸中隔墩岩体三个工程实例进行了免疫进化规划参数反分析、生态竞争模型岩体本构识别及灰色系统—进化神经网络位移序列建模预测研究,从工程实例角度验证了前述反分析方法的可行性。 计算智能在岩土工程中的应用研究是一个新兴的领域,本文仅进行了一定程度的计算智能反分析研究,但也未能穷尽。因此,本文研究只能起到抛砖引玉的作用,期望对此方面研究有兴趣的广大研究者共同努力,促进此项研究的发展。     

改进灰狼优化算法辨识光伏模型参数

针对标准智能优化算法辨识光伏(photovoltaic,PV)模型参数时存在准确度低、可靠性差和易出现早熟收敛的缺点,提出了一种改进的灰狼优化(improved grey wolf optimizer,IGWO)算法用于辨识PV模型参数。利用S型曲线的特点设计了一种基于S型函数的非线性调整控制参数a策略,以平衡算法的全局勘探和局部开采能力;以一定概率对当前最优决策层个体执行反向学习策略,帮助群体跳出局部最优。选取4个复杂函数测试IGWO算法的性能,利用实际光伏模型测量数据对IGWO算法进行检验。结果表明,IGWO算法相比其他算法,能更准确且稳定地辨识PV模型参数。     

基于频域介电谱的氧化锌阀片等效模型与参数辨识方法

目前对于金属氧化物避雷器阀片(ZnO阀片)的研究尚不全面,尤其缺乏针对其小电流区介电特性的研究.笔者依据ZnO阀片内部微观结构提出其介电等效电路模型,并通过测量频域介电谱(FDS)曲线的方法辨识该等效模型电气参数.本研究模拟避雷器阀片长时间交流老化的环境进行试验,并对不同老化程度的阀片样品进行FDS测量,使用粒子群算法与最小二乘算法相结合的智能融合算法对等效模型进行参数辨识,结果表明该融合算法辨识出来的模型参数能在1 mHz～5000 Hz范围内对实测FDS曲线进行高精度重构,从而达到描述其介电特性的目的.本研究为今后研究基于等效模型参数变化的避雷器老化受潮评价方法提供新的思路.     

岩石统计损伤软化本构模型及其参数反演

基于岩石微元强度服从幂函数分布的假定提出了损伤变量的一种新形式,由此建立了一个可模拟岩石破裂全过程的统计损伤软化本构模型.该模型含两个待定参数.其中一个可利用岩石三轴试验曲线的峰值点确定;另一个则受到微元强度参数的影响.通过研究微元强度参数对于软化本构模型的影响,提出了合理选择微元强度形式的参数反演方法,由此可最终确定模型参数.计算实例表明该模型能够有效模拟岩石的应变软化特性,应用反演方法确定微元强度参数能够有效提高数值模拟精度.     

一种改进人工蜂群算法的Wiener模型分步辨识法

为解决非线性系统Wiener模型参数辨识问题,提出了一种基于改进人工蜂群算法(NLJ-artificial bee colony algorithm,NABC)的分步辨识方法。NABC算法融合了人工蜂群算法的简洁性和随机搜索算法的高精度优势,有较快的收敛速度、较强的局部求精和全局搜索能力。通过对经典测试函数的寻优,验证了NABC算法较其他改进方法的优势。分步辨识法解决了Wiener模型辨识过程中容易造成的误差累积问题,实现其线性和非线性部分的独立辨识。通过pH中和反应过程建模仿真验证了分步辨识法的有效性。     

邓肯–张模型参数反演的不唯一性

对邓肯–张模型的特点进行了分析，指出待反演的参数中若同时包含破坏比R_f、凝聚力C和内摩擦角φ容易导致反演失败的原因在于反演结果不唯一所引起的数值奇异。证明了不唯一性源于参数R_f，C，φ之间的一种特殊相关性，即，对于任一特定的应力状态，有无数种R_f，C，φ组合可以保证R_fS（破坏比R_f与应力水平S的乘积）取值不变并获得相同的切线模量Et和切线泊松比ν_t，导致满足反演目标函数获得相同最小值的参数组合出现不唯一性。提供了相应的解决方法，即重新拟定反演参数或约束R_f，C，φ中的某些参数，以确保模型参数的可辨识性。     

一种冷水机组能耗模型参数的自适应在线辨识方法

本文针对冷水机组能耗模型参数辨识过程及冷水机组的渐变性及突变性,提出了冷水机组能耗模型参数的自适应在线辨识方法。首先,根据冷水机组能耗模型参数的变化情况建立了冷水机组的遗忘因子自适应模型,介绍了冷水机组能耗模型的递推最小二乘格式。而后,以某酒店离心式冷水机组为例对冷水机组能耗模型参数的辨识方法进行了实例分析。结果表明,变遗忘因子的自适应在线辨识方法在辨识过程中能够适应冷水机组能耗模型时变性的需求,提高了冷水机组能耗模型在系统渐变和突变过程中的辨识精度,为进一步的冷水机组在线运行优化控制及故障诊断提供了参考。     

微震源定位的两步反演方法研究

 微震源定位是微震监测技术的核心。微震源定位问题中的已知参数几乎都存在误差,现有微震源定位算法通常通过最小化所有检波器的到时(差)值函数,但得到的定位结果往往会偏离震源实际位置。针对现有微震源定位算法的不足,提出一种由两步反演构成的微震源定位新方法;在新方法中,一次反演识别异常检波器,二次反演实现震源空间坐标的精确搜索。基于均匀速度假设,分别建立3参数、4参数和5参数的反演模型。应用多目标遗传算法(Non-dominated Sorting Genetic Algorithm–II,简称NSGA–II)实现一次反演,为实现震源的精确搜索和减少计算时间,二次反演建议采用单目标优化算法实现。将提出的方法用于某矿井工程震源定位实例中,计算结果表明：提出方法能够有效的识别异常检波器,定位结果也较未剔除异常检波器时有了大幅度提升,且相对而言,4参数反演模型的定位结果优于5参数模型。该文方法可作为微震源定位的一种参考。     

岩石流变试验与本构模型辨识

岩石与岩体的流变性对结构、建筑物特别是大型水利水电和地下工程有很重要的影响。结合实践阐述岩石流变试验、模型辨识与参数反演以及流变模型的工程应用,提出岩石与岩体流变研究中若干亟待解决的重要问题。     

基于遗传算法的岩土本构模型辨识

通过对岩土本构模型辨识机理的分析，提出了基于遗传算法的本构模型辨识方法，并用两个工程实例对该法进行了验证。     

Numerical Simulation of Flow Failure of Geomaterials Based on Fluid Dynamics

A numerical method is developed for the prediction of large deformations associated with a geomaterial flow during the occurrences of such geo-disasters as landslides and slope failures. The geomaterial is modeled as a viscous fluid, where a Bingham type constitutive model is proposed based on Coulomb's failure criterion and the viscosity is derived from the cohesion and friction angle. Numerical experiments on a 2D gravitational flow of a geomaterial column show that the constitutive model performs well. For solving Navier-Stokes's equations, Constrained Interpolated Profile (CIP) scheme is utilized, which is able to efficiently treat solid, liquid and gas together and has been successfully applied to problems in fluid dynamics. To prevent numerical instabilities due to a large Bingham viscosity, an implicit calculation procedure is proposed to calculate the viscous forces in the CIP. The method is finally used to simulate an earthquake-induced landslide that took place on May 26th, 2003 in the Northern part of Miyagi Prefecture in Japan. Results from soil-slump tests on the samples from the landslide site are used for determining the parameters in the simulation. Good agreements between the calculated and the observed results are obtained.     

Digital representation of meso-geomaterial spatial distribution and associated numerical analysis of geomechanics: methods,applications and developments

This paper presents the author's efforts in the past decade for the establishment of a practical approach of digital representation of the geomaterial distribution of different minerals, particulars, and components in the meso-scale range (0.1 to 500 mm). The primary goal of the approach is to provide a possible solution to solve the two intrinsic problems associated with the current main-stream methods for geomechanics. The problems are (1) the constitutive models and parameters of soils and rocks cannot be given accurately in geomechanical prediction; and (2) there are numerous constitutive models of soils and rocks in the literature. The problems are possibly caused by the homogenization or averaging method in analyzing laboratory test results for establishing the constitutive models and parameters. The averaging method employs an assumption that the test samples can be represented by a homogeneous medium. Such averaging method ignores the fact that the geomaterial samples are also consisted of a number of materials and components whose properties may have significant differences. In the proposed approach, digital image processing methods are used as measurement tools to construct a digital representation for the actual spatial distribution of the different materials and components in geomaterial samples. The digital data are further processed to automatically generate meshes or grids for numerical analysis. These meshes or grids can be easily incorporated into existing numerical software packages for further mechanical analysis and failure prediction of the geomaterials under external loading. The paper presents case studies to illustrate the proposed approach. Further discussions are also made on how to use the proposed approach to develop the geomechanics by taking into account the geomaterial behavior at micro-scale, meso-scale and macro-scale levels. A literature review of the related developments is given by examining the SCI papers in the database of Science Citation Index Expanded. The results of this review have shown that the proposed approach is one of the latest research and developments in geomechanics where actual spatial distribution and properties of materials and components at the meso-level are taken into account.     

基于改进GA-SVR算法的隧道工程三维弹塑性模型参数的智能辨识

将支持向量回归（SVR）算法引入岩土工程数值计算模型参数的辨识中可以充分发挥SVR算法的小样本、泛化性好和全局最优化的优点。但现阶段标准的SVR算法只能解决一维输出变量的回归问题，这就使其在反分析领域的应用受到限制。引入一种改进的SVR算法，这种算法通过将多维输出变量回归转化为多层标准一维输出变量回归来解决这个问题，并与十进制编码的遗传算法相结合，形成改进的GA-SVR算法，用遗传算法搜索最优的SVR模型参数以建立最优的待辨识参数与位移之间的非线性映射关系，然后用遗传算法进行待辨识参数的最优辨识。为对比这种改进GA-SVR算法的效果，将遗传算法与BP神经网络相结合，形成GA-BP算法且编制相应的计算程序。将这两种算法运用于同样的隧道工程三维弹塑性模型参数的智能辨识，数值算例表明改进的GA-SVR算法较GA-BP算法可以取得更高的辨识精度和更好的计算效率，可运用于类似岩土工程计算参数的辨识。     

考虑屈曲影响的钢筋滞回本构模型

钢筋在拉压循环荷载作用下,会产生受压屈曲现象,其应力-应变曲线呈现受压应力软化效应,传统的钢筋滞回本构模型无法对其进行描述。在Bouc-Wen模型基础上进行修正,引入屈服后刚度比调整公式,建立考虑屈曲影响的钢筋滞回本构模型,并结合Newton-Raphson迭代法给出模型详细计算流程,该模型能够同时描述刚度退化、强度退化以及受压应力软化效应等钢筋滞回特性。模型滞回特性由7个基本参数进行控制,根据所收集的钢筋单轴拉压循环试验数据,采用遗传算法对其进行参数识别。之后考虑钢筋长细比、屈服强度以及强屈比的影响,对参数识别结果进行拟合,给出各控制参数计算公式。最后通过与试验结果进行对比,验证了所提模型的有效性。     

Thermo-hydro-mechanical-air coupling finite element method and itsapplication to multi-phase problems

In this paper, a finite element method （FEM）-based multi-phase problem based on a newly proposed thermal elastoplastic constitutive model for saturated/unsaturated geomaterial is discussed. A program of FEM named as SOFT, adopting unified field equations for thermo-hydro-mechanical-air （THMA） behavior of geomaterial and using finite element-finite difference （FE-FD） scheme for so/l-water-air three-phase coupling problem, is used in the numerical simulation. As an application of the newly proposed numerical method, two engineering problems, one for slope failure in unsaturated model ground and another for in situ heating test related to deep geological repository of high-level radioactive waste （HLRW）, are simulated. The model tests on slope failure in unsaturated Shirasu ground, carried out by Kitamura et al. （2007）, is simulated in the framework of soil-water-air three-phase coupling under the condition of constant temperature. While the in situ heating test reported by Munoz （2006） is simulated in the same framework under the conditions of variable temperature hut constant air pressure.     

Finite element model updating on small-scale bridge model using the hybrid genetic algorithm

Finite element (FE) model-based dynamic analysis has been widely used to predict the dynamic characteristics of civil structures. FE model updating method based on the hybrid genetic algorithm, by combining genetic algorithm and the modified Nelder–Mead's simplex method, is presented to improve bridge structures' FE model. An objective function is formulated as a linear combination of fitness functions on natural frequencies, mode shapes and static deflections using measurements and analytical results to update both stiffness and mass simultaneously. A commercial FE analysis tool, which can utilise previously developed element library and solution algorithms, is adopted for applications on diversified and complex structures. The validity of the proposed method is verified by using a simply supported bridge model with three I-shaped girders. FE models such as grid, beam-shell and shell model are considered to modify initial FE models on the experimental structure. Experimental results suggest that the proposed method can be applied efficiently to various FE models and is feasible and effective when this method is applied to identify FE modelling errors.     

Tank模型及其在边坡水位预测中的应用

 Tank模型可以模拟非线性的降雨–地下水运移过程,并且能迅速得到解答。基于现有的单列tank模型,提出新的复合水箱模型。由于新模型参数超过20个,应用传统优化算法难以快速找到最优解,一种新的启发式自搜索算法(变维数搜索算法)被引入并改进后用于模型最优解的寻找。变维数搜索算法能够根据搜索进程的变化自动改变搜索维数并且快速找到最优解。27个参数的复合tank模型被应用于日本国道九号线的一个边坡,计算结果表明：变维数搜索算法能够在10 min左右找到合适的最优解;降雨过程复合tank模型计算的地下水位变化和观测值非常接近。最后通过和有限单元法计算结果的比较表明,有限单元法的计算结果受地质渗透特性的影响很大,而复合tank模型不存在这种问题。工程实例计算表明,该方法和监测结果比较一致,但其适应性更强,特别适用于没有进行足够地质结构探查的边坡。它能够快速反映降雨过程中地下水位的运移过程,可以推广使用。     

Digital representation of meso-geomaterial spatial distribution and associated numerical analysis of geomechanics:methods,applications and developments

This paper presents the author’s efforts in the past decade for the establishment of a practical approach of digital representation of the geomaterial distribution of different minerals, particulars, and components in the meso-scale range (0.1 to 500 mm). The primary goal of the approach is to provide a possible solution to solve the two intrinsic problems associated with the current main-stream methods for geomechanics. The problems are (1) the constitutive models and parameters of soils and rocks cannot be given accurately in geomechanical prediction; and (2) there are numerous constitutive models of soils and rocks in the literature. The problems are possibly caused by the homogenization or averaging method in analyzing laboratory test results for establishing the constitutive models and parameters. The averaging method employs an assumption that the test samples can be represented by a homogeneous medium. Such averaging method ignores the fact that the geomaterial samples are also consisted of a number of materials and components whose properties may have significant differences. In the proposed approach, digital image processing methods are used as measurement tools to construct a digital representation for the actual spatial distribution of the different materials and components in geomaterial samples. The digital data are further processed to automatically generate meshes or grids for numerical analysis. These meshes or grids can be easily incorporated into existing numerical software packages for further mechanical analysis and failure prediction of the geomaterials under external loading. The paper presents case studies to illustrate the proposed approach. Further discussions are also made on how to use the proposed approach to develop the geomechanics by taking into account the geomaterial behavior at micro-scale, meso-scale and macro-scale levels. A literature review of the related developments is given by examining the SCI papers in the database of Science Citation Index Expanded. The results of this review have shown that the proposed approach is one of the latest research and developments in geomechanics where actual spatial distribution and properties of materials and components at the meso-level are taken into account. Translated from Chinese Journal of Rock Mechanics and Engineering, 2006, 25(5): 875–888 [译自: 岩石力学与工程学报]     

Numerical analysis of anisotropic stiffness and strength for geomaterials

In numerical modelling, selection of the constitutive model is a critical factor in predicting the actual response of a geomaterial. The use of oversimplified or inadequate models may not be sufficient to reproduce the actual geomaterial behaviour. That selection is especially relevant in the case of anisotropic rocks, and particularly for shales and slates, whose behaviour may be affected, e.g. well stability in geothermal or oil and gas production operations. In this paper, an alternative anisotropic constitutive model has been implemented in the finite element method software CODE_BRIGHT, which is able to account for the anisotropy of shales and slates in terms of both deformability and strength. For this purpose, a transversely isotropic version of the generalised Hooke's law is adopted to represent the stiffness anisotropy, while a nonuniform scaling of the stress tensor is introduced in the plastic model to represent the strength anisotropy. Furthermore, a detailed approach has been proposed to determine the model parameters based on the stress–strain results of laboratory tests. Moreover, numerical analyses are performed to model uniaxial and triaxial tests on Vaca Muerta shale, Bossier shale and slate from the northwest of Spain (NW Spain slate). The experimental data have been recovered from the literature in the case of the shale and, in the case of the slate, performed by the authors in terms of stress-strain curves and strengths. A good agreement can be generally observed between numerical and experimental results, hence showing the potential applicability of the approach to actual case studies. Therefore, the presented constitutive model may be a promising approach for analysing the anisotropic behaviour of rocks and its impact on well stability or other relevant geomechanical problems in anisotropic rocks.