A Multi-objective Optimization Evolutionary Algorithm with Better Performances on Multiple Indicators

Abstract

There exist conflicts among the performance indicators of Multi-objective Optimization Evolutionary Algorithm(MOEA), therefore, the design of such algorithms is also a multi-objective optimization problem. Currently, most MOEAs cannot obtain enough approximated Pareto front (APF) points, or have no enough approximation to true Pareto front, or have uneven point distribution or have incomplete coverage etc.. This paper proposes a new Multi-objective Evolutionary Algorithm to improve multiple performance indicators. This paper employs the Particle Swarm Optimization(PSO) with mutation operator and a multi-subpopulation strategy and let every subpopulation optimize a single objective to guarantee to visit the extreme points, uses a fast archiving algorithm with theoretical convergence to obtain enough points and good approximation, uses a crossover strategy among sub-populations and optimize a trade-off function to obtain trade-off solutions. Experimental results on eight widely used test-problems show that the performance indicators, including the numbers of front points, the uniformity, the complete of coverage and so on, are better than the compared algorithms: NSGA2, SPEA2, PESA etc. with satisfactory consumed time.