A Bayesian based Hyper-Heuristic approach for global optimization

Several metaheuristics have been developed for global optimization. Most of them are designed for solving a specific problem at hand, and their use on a new implementation is a challenging task. Hyper-heuristics are strategies that support these issues, combining a metaheuristic in a high-level for selecting or generating simple heuristics from a low-level. The aim is to find nearoptimal solutions based on the feedback received during the search. Estimation of Distribution Algorithms (EDAs) have been applied as hyper-heuristics, using a Probabilistic Graphical Model (PGM) to extract and represent interactions between its low-level heuristics to provide high-valued problem solutions. In this paper, we consider an EDA based on Bayesian networks as PGM on a hyper-heuristic context which encompasses a heuristic selection approach to find the best combinations of different known simple heuristics. We compare our proposed approach named Hyper-heuristic approach based on Bayesian Optimization Algorithm (HHBOA) using CEC'05 benchmark functions among 9 optimization algorithms. The experimental results show that HHBOA is competitive, outperforming the other approaches, especially in terms of convergence, on most of the functions considered in this paper.