Combining landscape approximation and local search in global optimization

Local search techniques have been applied in variant global optimization methods. The effect of local search to the function landscape can make multimodal problems easier to solve. For evolutionary algorithms, the usage of the step size control concept normally will result in failure by the individual to escape from the local optima during the final stage. We propose an algorithm combining landscape approximation and local search (LALS) which is designed to tackle those difficult multimodal problems. We demonstrate that LALS can solve problems with very rough landscapes and also that LALS has very good global reliability