关键词: 蒙特卡洛方法, 伪随机抽样, 准随机抽样, 拉丁超立方体抽样, 敏感性系数

Abstract: The Sobol method is one of the most important branch of sensitivity analysis. The produces of numerical solution of Sobol method mainly depend on the Monte Carlo method. While the generation of random numbers is the basis of Monte Carlo method. Pseudo random number generator is the most simple and basal method to generate random numbers. However, the uniformity of random numbers generated by pseudo random number generator is not very good. Thereby, the sample quality will be effected, and will affect the convergence and accuracy of sensitivity analysis indices calculated from the samples. In this paper, quasi random sampler and Latin hypercube sampler are used to substitute the pseudo sampler, then compare the samples get from them, and the accuracy and convergence of sensitivity analysis indices. Finally, the paper finds that quasi random sampling has obvious advantages compared with other two samplers.

Key words: Monte Carlo method, Pseudo random sampler, Quasi random sampler, Latin hypercube sampler, Sensitivity analysis indices