Researches Of Diversity Enhanced Particle Swarm Optimization And Its Application

Intelligent optimization algorithms provide an efficient solution to solve a class of dis-continuous, non-differentiable optimization problems. Particle swarm optimization (PSO) becomes one of the research hotspots in the field of intelligent optimization because of simple principle and realization, fewer tuned parameters and satisfactory convergence performance. Though PSO has been successfully applied in many fields, similar to other evolutionary al-gorithms, there still exist some drawbacks, such as premature and slow convergence speed because of quick loss of population diversity.In this paper, based on in-depth study of PSO, several modified PSOs were proposed to improve the convergence performance of PSO by enhancing population diversity. Meanwhile, simulation experimental results on a large number of nonlinear unconstrained optimization problems confirmed the convergence performance of the proposed modified algorithms. This thesis mainly includes the following research work:The first chapter mainly introduced the deficiencies of the traditional optimization meth-ods, the common swarm intelligence optimization algorithms, the state of the art of PSO and the main research work of this thesis.In the second chapter, simulating the disruption phenomenon, the disruption strategy was executed to those particles meeting the disruption condition during the search process through the corresponding disruption operator. The proposed different disruption operators have been applied to PSO and Bare-bones PSO (BPSO), respectively. In this paper, the author compared all kinds of the proposed disruption strategies, and compared the distribu-tion and population diversity of particles before and after executing the disruption strategy to illustrate that the disruption strategy can enhance population diversity. This strategy is beneficial to balance the global exploration and local exploitation abilities. Simulation exper-imental results and statistical analysis confirmed that the modified PSO algorithms (DPSO and DBPSO) can improve the convergence speed and precision.In the third chapter, the author changed traditional methods of opposition-based learn-ing (OBL), which is used only for population initialization or improving the population qual-ity, and utilized OBL to the personal best Pbest instead of the population X. On the one hand, this strategy provided a potential chance to update the global best Gbest, so it was beneficial to improve the convergence precision; on the other hand, the corresponding rela-tionship between Pbest and X was regrouped, which was conductive to re-guide the flying direction of particles. Meanwhile, the rebel learning item was integrated into the evolution e-quation to change the flying direction of particles, which was beneficial to enhance population diversity, and reduced the parameters by introducing random learning factors to decline the problems-solved dependence on parameters. The experimental results and statistical analysis confirmed that the modified PSOs (PSO-OBL and BPSO-OBL) can find better solution with smaller computation cost, and the dimension of problems-solved had less effect on them.In the fourth chapter, Inspired by the human learning behavior, the strategy learning from bad individuals (Gworst) was proposed. The value of learning factor(+,0,-), which obeys the standard normal distribution, was used to simulate the effect of bad behavior (impelled learning, not been affected, penalized learning). At the same time, two random learning factors (sum was 1) were employed to replace the two acceleration factors(c1 and c2) in PSO. These strategies reduced the parameters and changed the flying direction of particles, so it can reduce the dependence of the problems-solved on parameters and enhance the population diversity. The simulation experimental results and statistical analysis confirmed that the modified PSOs (HPSO) can improve convergence performance without increasing the complexity.