Improved Method of Multi-objective Particle Swarm Algorithm Learning Factor Based on Fitness Change

Abstract

Aiming at the problem that the inertial component of particles could not guide the particle to the right direction when the fitness became poor, a multi-objective particle swarm algorithm learning factor improvement method based on the fitness change was proposed. The large learning factor improved the multi-objective particle swarm algorithm. In the simulation experiment, the improved algorithm PSO-AIC1C2 and the PSO-S, PSO-AIC1 and PSO-AIC2 with c1and c2obtained by splitting this algorithm were fixed with c1and c2changed separately, and then compared with other PSO improvements. The algorithms MOPSO, SMPSO, and dMOPSO are compared. Experiments showed that increasing c1could improve the performance of the algorithm, and increasing c2would cause the convergence of the algorithm to deteriorate. In most test functions, PSO-AIC1C2 had obvious advantages in convergence and distribution indicators. The improved method proposed had certain guiding significance for the study of learning factors of particle swarm optimization in the future.