Hybridizing Cultural Algorithms and Local Search

In this paper, we propose a new population-based framework for combining local search with global explorations to solve single-objective unconstraint numerical optimization problems. The idea is to use knowledge about local optima found during the search to a) locate promising regions in the search space and b) identify the suitable step size to move from one optimum to others in each region. The search knowledge was maintained using a Cultural Algorithm-based structure, which is updated by behaviors of individuals and is used to actively guide the search. Some experiments have been carried out to evaluate the performance of the algorithm over several well-known continuous problems. The test results show that the algorithm can perform better than some current unconstraint numerical optimization versions of Cultural Algorithms in term of function evaluations and success rate in certain classes of problems.