Evolutionary continuous optimization by Bayesian networks and Guassian mixture model

In this paper, an evolutionary continuous optimization algorithm based on Bayesian networks and Gaussian mixture model (GMM) is proposed. A Bayesian network is used to model the relationship of variables in individual vector and the learned graphical structure is decomposed into subgraphs representing subproblems. Subsequently, GMM is adopted to model the probability distribution of each subproblem and its parameters are estimated by the expectation-maximization (EM) algorithm. New samples are generated from the GMM of each subproblem and Anally are mixed into new individuals. It is demonstrated by numerical examples that the proposed algorithm could achieve better performance than previous related algorithms.