An information theoretic criterion for adaptive multiobjective memetic optimization

Multiobjective memetic optimization algorithms (MMOAs) are recently applied to solve nonlinear optimization problems with conflicting objectives. An important issue in an MMOA is how to identify the relative best solutions to guide its adaptive processes. Pareto dominance has been used extensively to find the relative relations between solutions for the fitness assessment in multiobjective optimization based on evolutionary algorithms (MOEA). However, the approach based on the Pareto dominance criterion decreases its convergence speed when the number of objectives increases. In this paper, we propose an effective information-theoretic criterion based on the multiscale relative Rényi entropy to guide the adaptive selection, clustering, and local learning processes in our framework of adaptive multiobjective memetic optimization algorithms (AMMOA). The implementation of AMMOA is applied to several benchmark test problems with remarkable results.