Particle Swarm Optimization And Its Application

Particle swarm optimization algorithm (PSO) which was proposed by Eberhart andKennedy in1995is a new global optimization algorithm. PSO originated from the simulationof the behavior of bird predation, and used speed-location search model. The PSO algorithm,which is an efficient parallel optimization method, is used to solve a large number ofnon-linear, non-differentiable and multi-peak complex optimization problems. Procedure isrelatively simple and few parameters need to be adjusted, so the PSO algorithm developedrapidly and has been applied to the field of science and engineering. At the same time, manyimproved PSO algorithm appeared. Due to the relatively short history of the PSO, there arestill some problems need to be solved on the theoretical basis and its application.This paper focuses on the theory and application of PSO, especially, a systemic study onhow to improve the performance of PSO and analyze the trajectory of particles andconvergence. Solving the problems such as high dimension complex functions optimization,PID controller parameter tuning applications. The main research achievements can besummarized as follows:1Each parameter in the algorithm has been analyzed exhaustively; a single particletrajectory equation was given in one-dimensional space and multi-dimensional space. We alsoanalyzed the single particle’s convergence of the trajectory in the multidimensional space.2We proposed a late random inertia weight PSO (LRIWPSO)．Considering theshortcomings of particle premature convergence falling into a local optimum,(0.4,0.7)uniform distribution takes the place of the linearly decreasing inertia weight in search of thelate, Particles have larger w in the early stages of the search in order to maintain the diversityof particles and enhance the ability of global search.3We also proposed two kinds of inertia weight nonlinear dynamic adjustment PSO(NDIWPSO). The control factor m is introduced into the inertia weight of standard PSO, tocontrol w and the smoothness of the curvet, the algorithm is called NDIWPSO1. We alsoproposed the other non-linear dynamic inertia weight. In order to verify whether thesealgorithms have good performance, These algorithms are used to the optimization of the fourstandard test functions, The experimental results show that these algorithms can effectivelyovercome the inherent deficiencies of the particle swarm algorithm, avoid the prematureconvergence and improve the speed and accuracy of the search algorithm. The experimentalresults show that these algorithms are efficient global convergence algorithm.4Finally, NDIWPSO1algorithm is applied to the PID controller parameter tuning. Theexperimental results confirm the feasibility and effectiveness of the algorithm.