A hybrid multi-objective algorithm using genetic and estimation of distribution based on design of Experiments

In this paper, we design a hybrid multi-objective algorithm using genetic and estimation of distribution based on design of Experiments. At first, we apply orthogonal design and uniform design to generate an initial population so that the population individual solutions scattered evenly in the feasible solutions space. Second, we proposed a new convergence criterion to check whether the distribution of population has the obvious regularity. When the population is convergence, we use the model-based method to reproduce new individual solutions, otherwise genetic operator was employed to generate offspring. The results of systematic experiments show that the hybrid algorithm this paper proposed capable of finding much better convergence near the Pareto-optimal solutions and better spread of solutions than RM-MEDA.