Improving Imperialist Competitive Algorithm with Local Search for Global Optimization

Local search is commonly used in a population-based evolutionary algorithm to fine tune the quality of the solutions in the population. However, because local search is a costly process, previous work often suggests applying local search only to the current best solution instead of all of the solutions in the population. Imperialist Competitive Algorithm (ICA) is a population-based evolutionary algorithm that bases on the meta-heuristics of the human´s socio-political evolution. It divides its population into several sub-populations and allows these sub-populations to evolve independently and, at the same time, compete against each other. These subpopulations in ICA motivate the idea of applying local search on the best solution of each sub-population, as a compromise method between applying local search to all of the solutions and applying local search only to the best solution in the population. Our experimental results show that ICA with local search on the best solution of each sub-population yields competitive performance, when compared to ICA with local search only on the best solution of the population.