遗传算法在路面材料参数反演中的应用研究
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
随着我国交通事业的发展,公路无损检测与评价已显得越来越重要。如何根据无损检测设备检测得到的数据,反演路面结构层的模量,进而评价路面的承载能力在国际上开展了近三十年的研究,并取得了不少成果。遗传算法作为一种新兴的仿生优化方法具有许多优良特性,本文在遗传算法应用于路面反分析方面做了初步研究。
     遗传算法是模拟生物在自然环境中的遗传和进化过程而形成的一种自适应全局优化算法。它通过模拟达尔文“适者生存、劣者淘汰”的原理激励好的结构;通过模拟孟德尔的遗传变异理论在迭代中保持已有的优良模式,同时搜索更好的模式;通过群体的不断进化而搜索到问题的最优解。
     理论上遗传算法以概率1收敛于全局最优解,但在实际操作中单一的遗传算法往往存在早熟和局部最优点的现象。本文在系统介绍遗传算法基本原理的基础上,针对基本遗传算法收敛速度慢和早熟的现象提出了一些改进措施,并将其应用于路面反演问题;同时,针对遗传算法在路面反演具体问题中的补偿性规律,提出了在后期自适应调整模量搜索范围的改进方法。本文主要内容包括:
     1.针对遗传算法本身和遗传算法在路面结构反演中的具体问题提出了一些改进措施。
     (1)群体初始化以均布产生代替随机产生。
     传统遗传算法的初始化为随机产生,这样可能会漏掉一些好的基因模式,同时也可能造成多次对于搜索同一部位,造成机时浪费。本文初始群体的采用均布产生机制,使初始解均布在整个解空间内,从而,在很大程度上加快了收敛速度。
     (2)引入模拟退火的Metropolis接受准则,并采取回火退火策略。
     本文算法的总体框架遗传算法的框架,但对于遗传算法的变异结果,按照Metropolis准则判断接受与否。当温度较高时,以较大的概率接受恶化解,当温度趋于0时就不再接受任何恶化解了;但是,当温度趋于0时,可能还没有搜索到全局最优解,这时,可以人为地升高温度,使算法接受新解的概
    
     郑仆。1大学硕士学位论文 摘要
     率增大,跳离局部最优的陷阶,再重复进行杂交、变异、判断舍弃或接受、
     降温直至搜索到满意解。这个过程叫做回火退火。这样,由于算法增加了新
     的随机因素,就可以有效地防止遗传算法过早收敛或陷于局部最优。
    5(3)采取保优变异算子。
     即对群体中适应度最大的个体保持不变异,以避免群体退化。
     (4)自适应地调整后期模量搜索范围。
     算法进化若干代数后,解的质量相对较好时,根据当前群体的进化信息
     和模量反演的具体问题,自适应地调整模量的搜索范围,加强算法后期的局
     部搜索能力。
     2.基于系统识别基本原理和改进遗传算法,建立了路面反演的遗传算
     法,并用Fortran90编写了路面反分析程序。
     系统识别,就是根据系统的输入和输出数据来识别系统特性参数。它的
     基本思路是首先建立一个合理的模型来模拟末知系统,然后通过迭代过程来
     修改模型参数,使模型输出与实际系统输出之间的误差达到最小。对于路面
     反演问题,它的数学模型一般是
     l(X)二>.卜5(X)一1(l
     I二1
     式中,刀为弯沉盆的控制点数,一般为FWD设备传感器的个数。
     模量反演的遗传算法,就是将路面结构参数(通常为模量)作为决策变
     量,将其按照一定的方式编码为染色体,然后按照本文的改进遗传算法进行
     优化,反算结构参数。为了避免编码空间和解空间的相互转换,本文采用实
     数编码方案。
     基于以上原理,本文用Fo广ran90语言编写了路面反分析程序。
     3.利用本文的反演程序,分别对理论数据和实测数据做了数值分析。
     门)对理论数据做了稳定性分析。
     对于给定的路面结构,给以不同的初始值范围进行反演分析。数值结果
     表明,利用本文改进遗传算法反演的结果比较稳定,其结果基本不受初始值
     的影响。
     (二)对比分析了改进遗传算法和传统遗传算法的反演结果。
     一11-
     )
    
     郑州大学硕士学位论文 摘要
     一
     本文对SIDIP-LTPP模拟路面结构分别用本文的改进算法和传统遗传算
     法进行了反演。数值结果表明,改进遗传算法在求解质量和收敛速度方面都
     有一定程度的改善。
    二()对比分析了本文反演方法和国内外代表性的反分析软件。本文的反
     演
Along with the development of our country's pavement, nondestructive testing and evaluation is becoming more and more important. There are many achievements in backcalculating the pavement layer moduli and evaluating the pavement bearing capacity based on nondestructive testing data have been obtained in the last 30 years almost. Genetic algorithms(GA), as a rising biological modeling optimization method, have many good properties. Some preparative study about the application of GA in pavement backcalculation was made in this paper.
    GA is a selfadapting global optimization method, which is simulating the evolution procedure of biology in the environment. It can prompt good structures through simulating the Darwin's theory of "survival of the fittest, elimination of the poor". It can keep the good model and find the better model through simulating the procedure of the heredity and mutation of Mendelism, and finds the best one at the end of the population evolution.
    We can get the global optimal solution through single GA or SA without question in theory, but prematurity and local optimal solution is always met in operation of single GA or single SA. The fundamental theory of GA is introduced systematically in this paper. Some improvement measures are offered to eliminate the prematurity phenomenon and local optimal solution results. According to the compensatory property in pavement backcalculation, some improved methods are offered too. The main contents of this paper are as follows:
    1. Some improved methods are offered to the simple GA and to the application of GA in pavement backcalculation.
    (1 population generating uniformly instead of Population generating stochastically.
    
    
    
    Initializing population stochastically is always adopted in traditional GA, which will maybe omit some good models or generate some similar chromosomes, and then the unnecessary search will arise. Initializing population uniformly is used in this paper. With this method, the initial population is scattered in whole solution space, and the searching time can be reduced remarkably.
    (2) 'Metropolis' accepting rule of simulated annealing algorithm, backfire and annealing strategy are applied here.
    The overall frame of the algorithm in this paper is GA's frame, the mutation results should be judged according to the 'Metropolis' accepting policy. When the temperature is high, the deteriorative chromosomes are accepted at a high rate and will not be accepted in zero degree. If the temperature approximates to zero but the global optimum solution has not been gotten, then we should rise the temperature, and make the searching jump off the local optimum solution, and then redo the procedure of crossover, mutation, annealing, etc. until the global optimum solution is gotten. This procedure is named as backfire and annealing. Because new stochastic factor was introduced, the algorithm can avoid plunging into local optimum solution and prematurity phenomena.
    (3) Keep the fittest one in mutation operator.
    The best chromosome of the population is kept in mutation procedure to avoid population degeneration.
    (4) Adjust the searching area adaptive automatically.
    After some generation, when the results of evolution are good in a sense, according to the information of current population and the specific condition of GA in pavement backcalculation, the searching area should be automatically adapted. This measure can improve local searching ability.
    2. According to the fundamental of system identification and the improved GA, a backcalculation program has been developed with Fortran90.
    
    
    System identification is to identify system character parameter according to the input and output. The basic principle is to set up a reasonable model to simulate the unknown system at first, and then to modify the model parameter through iteration procedure to minimize the output data error between theoretical model and real system. For the pavement backcalculation, the mathematical model can be defined as follows:
    n- the number of control poi
引文
[1] 王复明,刘文廷.“八五”国家重点科技攻关项目高等级公路无损检测与CAE技术研究报告,路面结构反分析理论与方法研究.郑州工业大学,1996.
    [2] 王复明,刘文廷.高等级公路无损检测与评价—工程力学反问题研究.杨卫,郑泉水,靳征谟主编.走向21世纪的中国力学.北京:清华大学出版社,1996.230-238.
    [3] 查旭东.基于同伦方法的路面模量反算方法的研究.博士学位论文.长安大学,2001.
    [4] 周明,孙树栋.遗传算法原理及应用.北京:国防工业出版社.1999.
    [5] Bagley J D. The behavior of Adaptive System which Employ Genetic and Correlation Algorithm. Dissertation Abstracts International, 1967, 28(12).
    [6] Holland J H. Adaption in Nature and Artificial Systems. MIT Press, 1992.
    [7] De Jong K A. An Analysis of the Behavior of a Class of Genetic Adaptive Systems. Ph. D. Dissertation, University of Michigan, No.76-9381,1975.
    [8] Davis L D. Handbook of Genetic Algorithms. Van Nostrand Reinhold, 1991.
    [9] Goldberg D E. Genetic Algorithms in Search, Optimization and Machine Learning. Addison-Wesley, 1989.
    [10] Asphalt Pavement Overlay Design Guide, International Society for Asphalt Pavement, 1989.
    [11] A.H.A. Hogg, Equilibrium of a Thin Plate Synnetrucally Loded, Resting on an Elastic Subgrade of Infinite Depth, Phil. Mag., Set. 7, Vol. 25, March, 1938.
    [12] D.M. Burmister, The General Theory of Stresses and Displacements in Layered Soil System, Ⅰ, Ⅱ, Ⅲ, J. Appl. Phys., Vol. 16, No. 2, pp89-96; No.3, pp. 126-127; No.5, pp.296-302,1945.
    [13] 王伟.人工神经网络原理—入门与应用.北京:北京航空航天大学出版社,1995.
    [14] 查旭东,张起森.采用神经网络理论逼近三层体系弯沉的研究.“高速公路建设与发展”学术会议论文集.
    [15] 查旭东.基于神经网络理论的沥青路面结构四层体系弯沉的拟合研究.西安公路交通大学学报,2000,20(1):12-15.
    
    
    [16] 钟燕辉.路面结构层模量及路基深度反算方法研究[硕士学位论文].郑州工业大学,2000.
    [17] 王旭东,秦勤.一种动态弯沉盆的模量反算方法.公路交通科技,1999,16(1).
    [18] 孙立军,八谷好高,姚祖康.水泥混凝土路面板模量反算的一种新方法—惰性弯沉盆法.土木工程学报,2000,33(1):83-87.
    [19] 查旭东,张起森,王秉刚.基于同伦方法的路面模量反算方法的研究.全国第一阶旧路检测评价(罩面)及FWD应用技术研讨会论文集.
    [20] 王旭东,郭大进.落锤式弯沉仪模量反算的可靠性研究.中国公路学报,1999,12(3).
    [21] 王凌.智能优化算法及其应用.北京:清华大学出版社:施普林格出版社.2001.
    [22] 康立三,谢云,尤矢勇.非数值并行算法—模拟退火算法.北京:科学出版社.1998.
    [23] 谢云.模拟退火算法的原理及实现.高等学校计算力学及数学学报。1999,3:212-218.
    [24] 孙建国,马中高.用改进的模拟退火算法反演子波参数.石油物探.1998,87(3):77-81
    [25] 玄光南[日],程润伟.遗传算法与工程设计.北京:科学出版社,2000.
    [26] 王兴文.遗传优化设计方法及其在城市深基坑支护工程中的应用.博士学位论文.广西大学.2001.
    [27] 刘迎曦,王登刚,李守巨等.识别混凝土重力坝的一种新方法.大连理工大学学报.2000,40(2),144-147.
    [28] 孙建勇,申建中,徐宗本.一类自适应遗传算法.西安交通大学学报.2000,34(10),84-88.
    [29] SRINIVAS M, PATNAIK L M. Adaptive probabilities of crossover and mutation in genetic algorithms[J]. IEEE Trans Syst Man Cyber, 1994,24(4): 656-657.
    [30] 张文修,梁怡.遗传算法的数学基础.西安:西安电子科技大学出版社.2002.05.
    [31] EIBEN A E, AARTS E H, VAN HEE K M. Global convergence of genetic algorithms: a infinite Markov chain analysis[A]. SCHWEFEL H P, MANNER R H. Paral lei Problem Solving from Nature. Berlin: Springer-Verlag, 1991.
    
    
    [32] 陈章潮,顾洁,孙纯军.改进的混合模拟退火算法应用于电网规划.电力系统自动化.1999,23(10):28-40.
    [33] Fuming Wang, R., L., Lytton, System Identification Method for Backcalculating Pavement Layer Properties, Transportation Reserch Record, 1384.
    [34] 任瑞波,钟阳,张肖宁等.柔性路面结构参数反算的人工神经元法.哈尔滨建筑大学学报,2000,33(4):100-104.
    [35] 陈永兵,遗传算法及其在结构工程优化中的应用.西北工业大学硕士学位论文.2001.

© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号

地址:北京市海淀区学院路29号 邮编:100083

电话:办公室:(+86 10)66554848;文献借阅、咨询服务、科技查新:66554700