基于支持向量机的边坡可靠性分析

发布时间：2019-04-10 10:14

【摘要】：边坡可靠性涉及广泛的工程领域,是岩土工程研究的重要问题。传统的安全系数法无法考虑边坡中存在的各种不确定性因素,所得结果难以准确反映边坡的风险程度。本文将支持向量机法(SVM)与可靠性分析方法结合进行边坡可靠性分析方法与应用研究,主要内容如下： (1)针对某些复杂边坡功能函数无法显式表达,需借助软件计算安全系数的问题,建立边坡安全系数预测模型。首先采用均匀设计法选取一定数量的样本点,之后,通过确定性计算得到各样本点的安全系数,组成训练样本,训练SVM模型,拟合边坡的功能函数。实际算例表明,SVM模型能较好地拟合边坡的实际功能函数,准确高效地计算边坡安全系数。 (2)将SVM与一次可靠度法(FORM).二次可靠度法(SORM)、蒙特卡洛模拟(MCS)相结合进行边坡可靠性分析。采用FORM获得可靠度指标和验算点信息,通过迭代算法优化SVM模型,基于优化后的SVM模型,通过SORM和MCS计算边坡的失稳概率。用该方法对两个典型边坡进行可靠性分析,将结果与其他方法相比较,证明该方法具有准确性和高效性。 (3)基于SVM的可靠性分析方法所得结果能够反映参数敏感性及实际功能函数的非线性情况。通过FORM得到的U空间中的验算点能够反映参数的敏感性,通过SORM法得到的边坡极限状态面的主曲率K能够反映边坡极限状态面的非线性情况。 (4)通过对典型边坡的算例分析研究了参数分布类型、参数相关性、参数变异性对可靠性分析结果的影响。当不同分布函数的PDF曲线相似时,分布类型对可靠性分析结果基本没有影响,并且,PDF曲线差别越小,影响越小。边坡可靠度指标随参数变异系数的增大而减小,且参数敏感性越高,该变化越明显。参数相关性对可靠度指标的影响与参数相关性的正负性、极限状态面的形式以及参数敏感性有关。 (5)将随机场理论引入边坡可靠性分析,将插值法与SVM法结合对边坡进行可靠性分析,得到考虑参数空间变异性的可靠性分析结果。
[Abstract]:Slope reliability involves a wide range of engineering fields and is an important issue in geotechnical engineering research. The traditional safety factor method can not take into account all kinds of uncertain factors in the slope, and the results can not accurately reflect the risk degree of the slope. This paper combines the support vector method (SVM) and the reliability analysis method to study the slope reliability analysis method and its application. The main contents are as follows: (1) the function of some complex slopes can not be expressed explicitly. The prediction model of slope safety factor should be established by using software to calculate the safety factor. First, the uniform design method is used to select a certain number of sample points. Then, the safety factors of each sample point are obtained by deterministic calculation, the training samples are formed, the SVM model is trained, and the functional function of the slope is fitted. The practical example shows that the SVM model can fit the actual function of the slope and calculate the slope safety factor accurately and efficiently. (2) combining SVM with one-time reliability method (FORM). The reliability analysis of slope is carried out by the combination of (SORM), Monte Carlo simulation (MCS) and quadratic reliability method. The reliability index and check point information are obtained by FORM, and the SVM model is optimized by iterative algorithm. Based on the optimized SVM model, the instability probability of slope is calculated by SORM and MCS. This method is used to analyze the reliability of two typical slopes, and the results are compared with other methods. It is proved that the method is accurate and efficient. (3) the results of reliability analysis based on SVM can reflect the non-linearity of parameter sensitivity and practical function. The checking points in U space obtained by FORM can reflect the sensitivity of parameters, and the principal curvature K of slope limit state surface obtained by SORM method can reflect the nonlinear condition of slope limit state surface. (4) the influence of parameter distribution type, parameter correlation and parameter variability on reliability analysis results is studied by examples of typical slope. When the PDF curves of different distribution functions are similar, the distribution type has little effect on the reliability analysis results, and the smaller the difference of PDF curves is, the less the influence is. The slope reliability index decreases with the increase of parameter coefficient of variation, and the higher the parameter sensitivity is, the more obvious the change is. The influence of parameter correlation on reliability index is related to the positive and negative correlation of parameters, the form of limit state surface and the sensitivity of parameters. (5) the random field theory is introduced into the slope reliability analysis, and the interpolation method and the SVM method are combined to analyze the slope reliability, and the reliability analysis results considering the spatial variability of parameters are obtained.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU43

【参考文献】