Optimization in Continuous Domain by Real-coded Estimation of Distribution Algorithm

Finding global optima in the continuous domain is challenging for Genetic Algorithms (GAs). Traditional GAs use either binary-coded or real-coded representation of the variables, but there is a trade-off between these two encoding methods. Recombination operators for binary-coded GAs are simple to design, but the length of the string representing an individual would be huge if the number of design variables is larger; whereas, in real-coded GAs the length of an individual would be shorter, but the design of crossover and mutation operators are difficult, and sometimes they lead to premature convergence. To alleviate the problems of these two methods of encoding, real-coded Estimation of Distribution Algorithms (EDAs), which replace the recombination operators of GAs with estimation and sampling of the probability density function of the variables at each generation, have been proposed. In this paper, we show how real-coded EDAs can be applied to the optimization of multivariate functions in continuous domain and present the experimental results of three bench-mark functions produced by our proposed algorithm. In comparison with others EDAs, our proposed method obtains encouraging accuracy and efficiency.