Based On The Choice Of Goodness-of-fit Test Copulas Connect Function Estimation Of Distribution Algorithms

Estimation of Distribution Algorithm is a kind of stochastic optimizationalgorithm based on probability model, with the development and application oftechnology, it is more and more important to research how to select the properprobability model.In Estimation of Distribution Algorithm joint probability distribution isdecomposed into univariate marginal distribution and a copula. The study of themarginal probability distribution and the copula is relatively simple, and thecopula is easier to sample. In this paper we study the fitness between thepractical distribution and the joint probability distribution, analyse therelationship between the performance of Estimation of DistributionAlgorithm and the fitness of probability model.In this paper we study the relationship between the performance ofEstimation of Distribution Algorithm and the fitness of probability model.we select Archimedes copula as copula function, select normal distribution asthe univariate marginal distribution and estimate the parameter of copula basedon pseudo maximum likelihood. In Estimation of Distributed Algorithm theaccuracy of the probability distribution becomes very important, becausecopulas are different, and the corresponding relevance of the copula variablescould be different, inappropriate choice of probability model will make thesolution change for the worse. So we can adopt probability model that selectedthrough the test of goodness of fit to improve the algorithm. In this paper, wefirst introduce the Cramer-von Mises statistic Sn, then do simulation experimentbased on test of goodness of fit by using different null hypothesis and alternativehypothesis. The experimental results show that the method is effective. Secondin order to more accurately describe how the probability model of advantagepopulation influence the performance of algorithm, we adopt the method ofstatic adjustment of Archimedes copulas based on distance criterion, and theexperimental results show that the method is feasible. In order to more accurately describe the probability model of advantage population and cost lesstime as much as possible in building several probability models, At last weadopt use adaptive dynamic adjustment Archimedes copula function method, theexperimental results show the effective of this method, it can effectively improvethe performance of the algorithm.