Particle swarm optimization is a popular nature-inspired metaheuristic algorithm and has been used extensively to solve single- and multi-objective optimization problems over the last two decades. Several local and global search strategies, and learning and parameter adaptation strategies have been included in particle swarm optimization to improve its performance over the years. Most of these approaches are observed to increase the number of user-defined parameters and algorithmic steps resulting in an increased complexity of the algorithm. This paper presents a simplified multi-objective particle swarm optimization algorithm in which the exploitation (guided) and exploration (random) moves are simplified using a detailed qualitative analysis of similar existing operators present in the real-coded elitist non-dominated sorting genetic algorithm and the particle swarm optimization algorithm. The developed algorithm is then tested quantitatively on 30 well-known benchmark problems and compared with a real-coded elitist non-dominated sorting genetic algorithm, and its variant with a simulated binary jumping gene operator and multi-objective non-dominated sorting particle swarm optimization algorithm. In the comparison, the developed algorithm is found to be superior in terms of convergence speed. It is also found to be better with respect to four recent multi-objective particle swarm optimization algorithms and four differential evolution variants in an extended comparative analysis. Finally, it is applied to a newly formulated industrial multi-objective optimization problem of a residue (bottom product from the crude distillation unit) fluid catalytic cracking unit where it shows a better performance than the other compared algorithms.
Read full abstract