拉丁超立方抽样评估方法改进研究

Preliminary Study on Improved Latin Hypercube Sampling

为克服拉丁超立方抽样方法在均匀性方面的不足,对拉丁超立方抽样方法进行改进。引入均匀性评价参数G(L),给出了对于G(L)上限值的限定G_(max),并通过模拟试验的方式,对G_(max)与抽样维数N、样本数M进行了拟合,得到相应的函数关系。该方法满足了抽样均匀性的要求,并且可以保证样本均值与样本方差与原概率分布的均值与方差足够接近。

The uniformity in Latin Hypercube Sampling is still to be improved. To solve this problem, this paper presents an improved Latin Hypercube Sampling method. An evaluation parameter named G （L） for uniformity is introduced and upper limit of G（L） is defined as Gmax. With numerical simulation, a function of Gmax versus dimension degree N and sample number M is obtained by means of fitting curve. This method can meet the requirements of sampling uniformity and also ensure that the mean and variance are close enough to the original probability distribution.