The Research On Particle Swarm Optimization Algorithm By Using The Population Diversity Dynamic Adjustment Mechanism

Optimization problems exist widely in many fields of application such as the scienceresearch and economic management engineering and so on. The catholicity of optimizationproblems promotes development of optimization techniques and optimization algorithms.Optimization problems are becoming more and more complex with the rapid growth of theindustrial technology and scientific research, more and more researchers devote to proposehigh efficiency optimization algorithm.The particle swarm optimization could be well used to solve complex high-dimensionalproblems, especially for the problems that are lack of domain information. The particle swarmoptimization is concerned by many researchers since its principle is simple and its parametersare adjusted easily. As an intelligent algorithm based on group collaboration, the populationdiversity of particle swarm optimization is an important research focus. The individualbehavior and group trends are described by the population diversity which determines theevolutionary activity and convergence performance of the algorithm. From the perspective ofkeeping the population diversity, work on improving the performance of the populationdiversity has been done in this thesis. Detailed work and several new ideas are described asfollows:The affection of social structure on the population diversity is analyzed deeply,multi-ecology-subgroup social structure is used in this thesis. Each ecology-subgroup seeksextreme points in its way from its own environment which will slow down the speed ofparticle aggregation. Multi-ecology-subgroup social structure can hold the diversity directionduring the early process of the evolution.Two dynamic mechanisms on diversity are proposed, which are the full informationlearning strategy and the differential fusion strategy. The full information learning strategyuses the full-information particle as the bridge of communication among particles, and thefull-information particle contains all of the useful information of particles in oneecology-subgroup. During the evolution process, the parameters of the full-informationparticles are calculated dynamically to adjust the combined ratio of particles and supply morechoice for particles learning. The dynamic character of it is good for the diversifieddevelopment in the direction of particle motion. The differential fusion strategy considers thechange trend of the populations as the identification. The differential fusion strategy inspectsand adjusts the social structure to controls the diversity direction during the evolution processdynamically. The proposed algorithm is tested on the classical test functions and the real logisticsproblems to test its performance, the experiment results show that the new algorithm has goodperformance of searching for optimization, and it is also has strong practicability for the realproblems.