Multi-Objective Social Group Optimization with Dynamic Fitness Function and Crowding Distance Elimination
TL;DR
Experimental results show that the proposed MOSGO provides, in most problems, significantly better convergence near the true Pareto front, with improved diversity and spread of solutions, compared to other multi-objective algorithms.
Abstract
Many real-world optimization problems involve conflicting objectives that need to be minimized to reduce cost and/or maximized to increase profit. In this study, a multi-objective social group optimization (MOSGO) is proposed and implemented to solve multi-objective problems and find approximated solutions to the optimal Pareto front. A new mechanism, the dynamic fitness function, is introduced and integrated with non-dominated sorting and crowding distance elimination strategies to enhance the quality of the non-dominated solutions. The dynamic fitness function is designed to select the best solution for each objective at each iteration. Non-dominated sorting is used to dismiss weak solutions, and crowding distance elimination is deployed to achieve the best solution diversity. The suggested algorithm is compared with four competitive algorithms: the multi-objective artificial hummingbird algorithm (MOAHA), the multi-objective particle swarm optimization (MOPSO), the multi-objective ant lion optimizer (MOALO), and the non-dominated sorting genetic algorithm-II (NSGA-II). Computational simulations are performed on well-studied ZDT benchmark test functions. Comprehensive comparisons are carried out regarding convergence, diversity, and solution distribution. Experiment results show that the proposed MOSGO provides, in most problems, significantly better convergence near the true Pareto front, with improved diversity and spread of solutions, compared to other multi-objective algorithms.