Based On K - S Inspection Copulas Connect Distributed Estimation Algorithm To Study The Distribution Of The Edge

Estimation of distribution algorithm (EDA) is a new evolutionary pattern,which is developed on the basis of genetic algorithm(GA), but it is differentbetween the two algorithms that the former is through the establishment ofprobability model to describe the dimension distribution instead of crossoverand mutation operation of the genetic algorithm, and using statistical learningmethod to analyze dominant group and establishing the correspondingprobability model, then sampling from the probability model is to produce newindividuals, adding it into the population of the next generation, so again, it isnot stop until satisfy a termination condition,realizing the evolution of thepopulation.Estimation of distribution algorithm based on copula (copula EDA),combined Copula theory with estimation of distribution algorithm, which showsboth the advantage of the copula theory and estimation of distribution algorithm,meanwhile making up the weaknesses of the estimation of distributionalgorithm. Copula theory divides joint distribution which estimates by dominantgroup into two parts: the estimation of the marginal distribution functions andthe estimation of copula function and sampling. Apparently the estimation of themarginal distribution and the copula function and sampling are much simplerthan to directly estimate the joint distribution function, thereby reducing thetime overhead, and increasing the estimation accuracy, as well as improvingexecution efficiency.The selection of edge distributions in copula EDA are varied, and differentselection strategies have a great influence on optimization result.This paper isarranged, first of all, selecting Clayton copula as connection function andselecting experience distribution and Cauchy distribution as the margindistribution functions, it is to verify the feasibility and validity of the Cauchydistribution as edge distribution function. And then analyzing the optimizationresults are analyzed and compared, it is found that the optimization results ofCauchy distribution as marginal distribution function are much better, but it isstill some premature phenomenon to some kinds of test function in optimization results.The marginal distribution of copula estimation of distribution algorithm isbeen further analysis that the edge distribution function adopting only onedistribution function (such as the experience of distribution, normal distribution,Cauchy distribution etc) ignores random and diversity of the sampledistributions. Accordingly, a variety of different distribution functions includethe normal distribution, Cauchy distribution, t distribution are simultaneouslyapplied to the candidate margin distribution functions, and using K-S test todetermine the distribution of each dimension, each dimension variable may havetheir own distribution function, there can be the same or different, if onedimensional distribution function can not find from the candidate the marginaldistributions, then constructing the corresponding empirical distributionfunction, According to the simulation experiment, the thought of the copulaEDA based on K-S test which selects the margin function has a better optimalvalue and can quickly converge to the optimal value, the population of laterperiod has also relative stability.