Particle Swarm Optimization (PSO) algorithm is an important branch of swarmintelligence. Due to its simple concept, simplicity of implementation, less parameterto control and rapid convergence speed. Once it has aroused widespread concern, andimproved by many researchers to investigate. PSO algorithm has been widely appliedin both theoretical and practical fields, such as to combination optimization, networkoptimization and production scheduling etc.This paper presents a detailed overview of the basic concepts of PSO algorithm.By analyzing the principle of the algorithm, it gives out some improved versions toavoid the premature convergence problem. Particle swarm optimization algorithm andthe current research status of its improvements and applications are investigated. Inthis paper, some improved algorithms are proposed, the content of as follows:1) When PSO algorithm is applied to solve multimodal problems andcomplicated optimization problems with local optimum, the best particle in thealgorithm will cause particles convergence to the local optimum, which will lower theperformance in convergence speed and search ability. To overcome this problem, animproved algorithm is proposed that Particle Swarm Optimization based onself-adaption of step size u with the average value (SAUPSO). For a certaindimension of the particle with better performance, the velocity step will be smallerand the search range will be more accurate. Not only is the swarm average adaptationvalue influence considered for the selection probability in the step factor selection butalso the individual adaptation value. The aim is to enable the individual that has somepoor adaptation, but good evolution tendency to preserve and use. The analysis of theSAUPSO algorithm performance simulated by experiment and results show theproposed approach has better convergence speed and search ability than PSOalgorithm.2) Most bats produce echolocation sounds by contracting their larynx. Put theecholocation strategy into the Bat Particle Swarm Optimization (BAPSO) algorithm.Give each particle is different frequency, what was used to determine the searchingrange, then a detailed search for regional around the optimal solution must be donewith pulse rate. The results on five benchmark functions prove the feasibility andvalidity of this method.3) In order to avoid the premature convergence in the Chaotic Particle Swarm Optimization (CPSO) algorithm, a Mutation Chaotic Particle Swarm Optimization(MCPSO) algorithm is proposed. A different variation coefficient instead of the sameone for each particle to decide when to variation. In addition, referring to use theGauss mutation and the uniform mutation to variety the location of each particle inorder to confirm that these location would not change nothing or change very little ina long time, so that the search ability can be preserved and the ability of the avoidingthe premature convergence will be improved, and has the more probability to attainthe best value. In the other hand, to make use of different advantage around two typesof variation, in order to balance the global and local search. Experiments onbenchmark functions show the improvement on the MCPSO algorithm is better thanCPSO algorithm. Stochastic convergence criterion is used for the convergence ofMCPSO algorithms.4) Application of PSO algorithm in discrete space, in recent years has beenwidespread concern, especially with it to solve combinatorial optimization problem.This paper applies the three kinds of improved PSO algorithm in this paper to solveknapsack problem, and separately using two simulation experimental data, the datashow that, compared with PSO algorithm, improved PSO algorithm can converge tothe optimal solution, and the convergence speed is improved. |