Extension method for Latin hypercube sampling

Abstract

The present invention discloses an extension method for Latin hypercube sampling. An LHS (Latin hypercube sampling) of which a sample size is n is supposed to exist, the extension method aims to obtain a new LHS sampling of which the sampling size is n+m, and meanwhile, the novel sampling contains a maximum number of existing sampling points. According to the method provided by the present invention, a relationship between the LHS sampling points is transformed into a simple undirected acyclic graph, an adjacent matrix is obtained, and a maximum individual set is solved to retain a maximum number of existing sampling points. The method comprises the implementation steps of: calculating distribution of the existing sampling points in a novel sampling structure; calculating a matrix of the sampling points that need to be retained; calculating a matrix of newly added sampling points; and finally, calculating a Latin hypercube sampling matrix after extension. According to the extension method for the Latin hypercube sampling, which is provided by the present invention, the existing sampling points can be deleted as few as possible, so that sampling cost is conserved; and meanwhile, the novel sampling points are generated, so that a sampling sample formed by the novel sampling points and the retained sampling points still meets an LHS structure.

Description

technical field [0001] The invention relates to the field of data information processing, in particular to an expansion method of Latin hypercube sampling. Background technique [0002] Latin hypercube sampling (LatinHypercubeSampling, LHS) is a full space filling and non-overlapping random sampling method. Full space-filling sampling ensures that information about the entire design space of the source function can be obtained even without detailed source function characteristics; non-overlapping sampling ensures that there are no repeated and redundant sampling points. LHS is also a random sampling method, but unlike random sampling, the sampling points it generates are uniform globally. The traditional Latin hypercube sampling algorithm is based on a fixed number of samples. In the application, it is necessary to construct a fixed sampling structure and sampling points according to the predetermined number of samples, and then calculate the result samples; if the calculat...

AI Technical Summary

Problems solved by technology

In the traditional extension method, the new sampling cannot be guaranteed to be LHS, because the change of the sample size leads to the change of the LHS sampling space structure, and the existing sampling points may not satisfy the LHS structure, such as figure 2 As shown, some of the existing sampling points fall in the same interval in the new structure, which is no longer strictly Latin hypercube sampling

Images