【作者】 纪华；

【导师】 郑璐石；

【作者基本信息】 宁夏大学 ， 固体力学， 2005， 硕士

【摘要】 支持向量机(SVM)是基于统计学习理论的一种新的机器学习方法。由于具有较完备的理论基础和较好的学习性能,能很好地解决小样本、非线性、高维数和局部极小点等实际问题,使它成为继神经网络之后新的研究热点。尽管SVM方法的性能在很多实际应用中得到了验证,但在计算上仍存在一些问题,包括训练速度慢、算法复杂而难以实现、对噪声和野值敏感及算法实现效率低等。 岩土力学是一门既赋理论内涵,又工程实践性很强的学科。经过几十年的研究,运用材料力学、弹性力学、弹塑性理论等传统科学,提出了各种求解方法,但由于岩土介质的复杂性、非线性、不确定性和模糊性等特点,传统的方法并没有达到满意的效果。智能岩石力学的提出和发展为其提供了一条崭新的途径,SVM作为一种新的通用机器学习方法,对解决很难用数学模型描述的问题具有很好的适应性,并具有广泛的应用前景。 在岩土工程问题中,如何正确给定岩体力学参数一直是一个比较棘手的问题,反分析方法为岩体力学参数的获取提供了较为有效的途径。本文按照智能岩石力学的思想,将SVM应用于岩土力学反分析的研究中,对岩土参数的识别进行了研究。 论文的主要工作有: 1.概述了统计学习理论中关于小样本统计的部分重要结论,详细地介绍了SVM的基本原理、算法、特点以及存在的问题,并讨论了它与统计学习理论中相关结论的关系; 2.讨论了SVM存在对噪声和野值敏感的问题,分别介绍和分析了目前针对此问题所提出的一些方法,在此基础上提出了一种改进的支持向量机算法,并用人工合成数据验证了算法的可行性和有效性; 3.针对SVM在参数(包括核函数及其参数)确定方面的问题,提出了基于模拟退火算法的SVM方法,该方法既利用了模拟退火算法的全局优化能力,又利用了SVM在处理小样本、高维数、非线性等问题方面的优良特性; 4.将SVM引入到岩土工程研究中,提出了岩土体参数识别的模拟退火支持向量机方法,并通过算例说明了该方法的可靠性。更多还原

【Abstract】 Support Vector Machine is a kind of new machine studying method, which is based on Statistical Learning Theory. Because it has quite perfect theoretical properties and good learning performance, and can solve some practical problems such as a little sample, non-linear, high-dimension and part minimized value, SVM becomes the new research hotspot after the research of Artificial Nerve Net. However, SVM performance has been validated in many practical applications, there are still some drawbacks. For example: train speed is slow, algorithm is complex, SVM’s implementation is efficient and it’s adaptability to noises and outliers.Geomechanics is a subject with the deep theory and strong practice. Through a few decades, a lot of methods have been applied to solve the problems on geomechanics and elastic and plastic theory. But owing to rock and soil’s features such as the complexity, non-linearity, randomty, uncertainty and obscurity, it is difficult to get satisfactory results by traditional ways. The development of the intelligent rock mechanics offers a new method. SVM is a new method of machine learning, it is fitful for the problem that can’t be solved with traditional mathematical model. There is an extensive prospect in geotechnical engineering.In the solution of Geotechnical, it is intractable to ascertain parameter of rock and soil exactly. Back analysis is a good method to get parameter of rock. In this paper, the author applied SVM to geomechanics, and studied the parameter identification of rock and soil. The author’s major works are as the following:1 The author briefly summarized some important conclusions of Statistic Learning Theory and discussed in detail the principle of SVM in pattern recognition.2 The author mainly discussed the problems such as the efficiency in SVM’s implementation and it’s adaptability to noises and outliers. Recent research on these problems and several algorithms are introduced and analyzed in the overview. According to the discussion, we proposed a modified SVM. Some experiments of its performance on artificial data are presented to show its feasibility and effectiveness.3 The author proposed a new method—support vector machine’ based on simulated annealing algorithms, which combines the global optimization characteristic of simulated annealing algorithms and the nonlinear mapping characteristic of support vector machine. *4 The author applied the support vector machine to geoengineering, and proposed the simulated annealing—support vector machine method to recognize the parameter of rock and soil, a example prove this method is right.更多还原