中文摘要：

抗剪强度参数最优概率分布形式的推断是保证岩土工程可靠度计算结果精确的基础和前提。现有研究认为大多数抗剪强度参数服从正态或对数正态分布,但是由于岩土参数实际分布区间有限,导致上述分布在使用过程中均存在和实际参数分布区间不匹配的问题。考虑到绝大多数岩土参数存在偏度这一事实,提出以"3?"原理为基础,并考虑偏度进行调整的分布区间确定方法。并基于正态信息扩散原理,提出推断岩土抗剪强度参数概率分布的正态信息扩散法。以水利水电工程中的3组岩基内摩擦角样本作为实例,利用正态信息扩散法推断对应的概率密度函数,并以K-S法进行检验。同时为了考察样本容量对正态信息扩散法以及经典分布拟合法在拟合精度方面的影响,利用蒙特卡洛模拟方法生成已知分布的8组随机样本,样本容量分别为15,20,30,50,100,200,500和1000。研究结果表明:不论对实测样本还是模拟样本,与传统的经典分布拟合法得到的最优分布——对数正态分布相比,正态信息扩散法的检验值均低于对数正态分布,并随着样本容量的增加逐渐趋于收敛,而对数正态分布检验值的收敛趋势相对弱一些。另外,在文中确定的分布区间内,正态信息扩散分布累积概率值的误差约在10-4量级,并且几乎不受样本容量变化的影响;对数正态分布累积概率值的误差在10-3量级,但是受样本容量变化的影响很大,并呈现不规则变化趋势。

英文摘要：

The inference of optimal probability distribution of shear strength parameters is the basis and premise to ensure the accuracy of reliability calculation in geotechnical engineering. The existing studies suggest that most of the shear strength parameters obey the normal or logarithmic normal distribution. However, because the actual distribution range of geotechnical parameters is very limited, the problem that range mismatches between the defined interval of normal distribution or logarithmic normal distribution and the actual distribution interval of geotechnical parameters is inevitable. Considering the fact that there is a certain degree of skewness for the distribution of most geotechnical parameters, based on the "3s " principle, a distributed interval determination method adjusted with the skewness is proposed. Three groups of samples of the internal friction angle of batholiths from water conservancy and hydropower projects are treated as examples, and the normal information diffusion method (NID method) is used to infer their respective probability distribution function. The K-S test method is also introduced to test the fitting degree. At the same time, in order to investigate the influence of sample sizes on the fitting accuracy of the normal information diffusion method and the typical distribution fitting method, eight groups of samples are produced using the Monte-Carlo method, and the sample size is 15, 20, 30, 50, 100, 200, 500 and 1000. The results show that, regardless of the actual or simulated samples, compared with the logarithmic normal distribution (obtained by the typical distribution fitting method), all the test values of the normal information diffusion distribution are lower than those of lognormal distribution, and tend to converge with the increase of the sample sizes.