岩土参数概率分布的系统推断研究

Study on System Inference of Probability Distribution of Geotechnical Parameters

借助数据挖掘技术,将岩土参数测试样本概率分布的选择系统化。选取了19组岩土参数测试样本,选择beta分布和正态信息扩散法作为推断岩土参数概率密度的两种方法。借助数据挖掘技术中的模糊C-均值聚类方法,根据岩土参数测试样本的数字特征值,将19组岩土参数测试样本归聚为4类,再根据每一类样本的分布规律,确定其概率密度函数的推断方法。对于波动性和离散性严重的测试样本,在采用正态信息扩散法推断其概率密度函数时,为解决窗宽选择问题,提出以均方差最小为目标,在窗宽的基础上,改变窗宽的大小,确定样本最优窗宽的方法。对于一组新的测试样本,根据数字特征值即可推断出测试样本的概率分布。

With the help of data mining technology,the selection of probability distribution of geotechnical parameters for test samples was systematized.19 groups of geotechnical parameters for test samples were selected,the beta distribution and the normal information diffusion method were chosen as two methods to infer the probability density of geotechnical parameters.Based on the fuzzy C-means clustering method in data mining technology,19 groups of geotechnical parameters for test samples were clustered into 4 categories according to the numerical eigenvalues of samples,and then the inference method of probability density functions was determined according to the distribution law of each category for samples.For test samples with serious fluctuation and discreteness,an optimal determination method on window width was proposed,the method took the minimum mean square error as the constraint goal and changed the size of window width on the basis of window width.When the probability density function was deduced by the normal information diffusion method,in order to determine window width,an optimal determination method on window width of samples by changing the width value of the window on the basis of the minimum mean square error was proposed.For a group of new test samples,it was probable to infer the density function of samples by the numerical eigenvalue.