A Fast Multi-Objective Particle Swarm Optimization Algorithm Based on a New Archive Updating Mechanism

Wenli Du,Khalil Alkebsi

doi:10.1109/access.2020.3007846

Abstract

Multi-objective optimization has received increasing attention over the past few decades, and a large number of nature-inspired metaheuristic algorithms have been developed to solve multi-objective problems. An external archive is often used to store elite solutions in multi-objective algorithms. Since the archive size is limited, it must be truncated when the number of nondominated solutions exceeds its maximum size. Thus, the archive updating strategy is crucial due to its influence in the performance of the algorithm. However, achieving a fast convergence speed while assuring diversity of the obtained solutions is always a challenging task. In this paper, a novel multi-objective particle swarm optimization algorithm based on a new archive updating mechanism which depends on the nearest neighbor approach, called MOPSONN, is proposed. Two archive updating strategies are adopted to update nondominated solutions in the archive, which are beneficial to accelerate the convergence speed and maintain diversity of the swarm. The performance of MOPOSNN is evaluated on several benchmark problems and compared with seven state-of-the-art multi-objective algorithms, including four multi-objective particle swarm optimization algorithms and three multi-objective evolutionary algorithms. The experimental results demonstrate the significant effectiveness of MOPSONN in terms of convergence speed and spread of solutions.

Highlights

In many real-world optimization problems, the problem involves two or more conflicting objective which need to be optimized simultaneously [1]–[3]
To further improve the convergence speed and diversity preservation of particle swarm optimization (PSO) in solving multi-objective optimization problems (MOPs), we propose a novel variant of multi-objective PSO (MOPSO) based on a newly developed archive update mechanism, termed MOPSONN
In this paper, a novel multi-objective particle swarm optimization MOPSONN based on a new archive updating mechanism is presented

Summary

INTRODUCTION

In many real-world optimization problems, the problem involves two or more conflicting objective which need to be optimized simultaneously [1]–[3]. As most of dominationbased multi-objective optimization algorithms make use of infinite external elitism archive to store nondominated solutions found so far, the update mechanism of this archive is crucial. To further improve the convergence speed and diversity preservation of PSO in solving MOPs, we propose a novel variant of MOPSO based on a newly developed archive update mechanism, termed MOPSONN. Similar to vicinity distance update procedure, all solutions whose crowding distance is changed after removing a solution need to be updated by calculating their distance to their new closest solutions By applying these two rules, we can achieve better balance between convergence and diversity. 3) A novel multi-objective particle swarm optimization algorithm, called MOPSONN is proposed based on a new archive updating procedure and pairwise competition mechanism. Where ω is the inertia weight, t is the generation number, c1 and c2 are the learning factors of the personal best position and global best position, respectively, and r1, r2 ∈ [0, 1] are two random numbers [32]

EXISTING MOPSO ALGORITHMS

THE COMPLETE MOPSONN ALGORITHM

CONCLUSION