Non-autocorrelation sampling method for large sample space complex probability distribution

Abstract

The invention discloses a non-autocorrelation sampling method for large sample space complex probability distribution, which aims to solve the autocorrelation problem of an MCMC method and the efficiency loss problem of skip sampling. According to the technical scheme, when samples are generated, the correlation among the samples is eliminated by setting a sample buffer area; or as a post-processing method, the generated samples are read in sequence, and the sample sequence is updated through the sample buffer to eliminate the autocorrelation of the sample sequence. When one sample is generated each time or one sample is read from the file, the obtained sample is stored into the sample buffer area, and the sample is randomly selected from the buffer area to output when the buffer area is full. By means of the random output mode, a layer of extra randomness is added between samples, and finally the purpose of eliminating the sample sequence autocorrelation is achieved. By adopting the method, the calculation speed is fast, the MCMC efficiency is maintained, the expenditure of the additionally-increased buffer region operation can be ignored relatively, and meanwhile the sample autocorrelation is eliminated.

Description

technical field [0001] The invention relates to a sampling method for complex probability distribution samples in a large sample space, in particular to a Markov chain-based non-autocorrelation sampling method. Background technique [0002] Computer processing of random events is a necessary means of engineering application analysis. In recent decades, with the rapid development of computer performance, computers have been widely used in aerospace, automobile and ship design and manufacture, bridge building design and manufacture, environmental engineering, weather forecast, polymer materials, etc., and in these engineering applications, Both need to calculate and analyze random events through computers to achieve purposes such as bridge reliability testing and weather forecasting. In random event analysis, a necessary step is to sample complex probability distributions to obtain independent, random samples. [0003] An efficient sampling algorithm is a necessary means to ...

AI Technical Summary

Problems solved by technology

[0065] The technical problem to be solved in the present invention is aimed at the autocorrelation problem of the MCMC method and the efficiency loss problem of skip sampling, and provides a kind of non-autocorrelation sampling method for the complex probability distribution of the large sample space, so as to achieve the same sampling rate as the MCMC method. Under the premise of efficiency, generate a sample sequence without autocorrelation

Images