MACO/NDS: Many-objective Ant Colony Optimization based on Non-Dominated Sets

A new model for many-objective optimization based on ant colony is presented. It decomposes a domain with several objectives into subdomains with pairs, triples and n-tuples of objectives. Pheromone structures related to each subdomain guide the ant population in the construction of non-dominated solutions for the different multiobjective sub-problems, increasing the convergence to the Pareto Optimal of the original problem. The new algorithm was called Many-objective Ant Colony Optimization based on Non-Dominated Sets (MACO/NDS). The proposed model was evaluated in two discrete challenges: the multiobjective knapsack problem (MKP) and the multicast routing problem (MRP). The performance of the new model was confronted with the widely known many-objective algorithms MOEA/D and NSGA-III, in addition to the evolutionary models MEAMT and MEANDS proposed for discrete optimization problems that have also been evaluated in MKP and MRP. The results show that the MACO/NDS is competitive using instances with 4-6 objectives, proving to be a stable algorithm with good results in both problems.