A novel framework for multi-objective optimization problems

Abstract

Even though multi-objective optimization problems’ solution space is inherently complex, heuristic based algorithms often search the solution space by a single neighbour creation technique. Dynamic neighbour generation (DNG) allows searching solution space by multiple heuristic operators and brings a new perspective to neighbour creation process especially for the multi-objective optimization problems. This paper presents extensive comparative experiments for the purpose of analyzing and revealing the achievement of our proposed DNG framework on a set of benchmark problems. DNG is integrated with the fast and elitist multi-objective genetic algorithm (NSGA-II), multi-objective migrating birds optimization algorithm (MMBO), the strength Pareto evolutionary algorithm 2 (SPEA2) and Pareto simulated annealing (PSA). Multi-objective hyper-heuristic evolutionary algorithm (MHypEA) is also implemented for obtaining a more effective comparison. Experiments demonstrate that DNG based versions of algorithms are 24.39% and 28.35% better on the average than their original variants in terms of inverted generational distance indicator and in terms of hypervolume indicator respectively.