Journal Homepage: www.ijcst.org

Abstract – Evolutionary Algorithms (EAs) represent an elegant class of solution paradigms that can efficiently tackle NP-hard problems such as network design problems. The most widely used of these EAs is genetic algorithm (GA). However, GA is prone to premature convergence making it unable to search numerous solutions of the problem domain. A memetic algorithm (MA) which is a symbiosis of GA and local search technique is an effective option for reducing the likelihood of premature convergence. This paper proposes a MA-based approach for multi-objective design of communication networks. To be able to estimate the quality and cost (in computation time) of obtained MA solutions, we design a GA and use it to equally solve the problem. Our computational experiments show that MA is superior to GA in solution quality but inferior to GA in computational time. However, for 36-node network (large network), MA is able to find solutions (though not best) that are better both in quality and cost than the best GA solution