Improved Particle Swarm Optimization Algorithm And The Application Of Artificial Neural Network

Particle Swarm Optimization (PSO) is a new kind of evolutionary computation and was originally introduced by Kennedy and Eberhart in 1995.Because of its simple concept and easy implementation with only a few tuning parameters, it has been attracting more and more attention. It is used in many fields such as function optimization, neural network training, fuzzy system control and other applied fields.First, the paper analyses the important meaning of giving a comprehensive study on PSO, then, several aspects of PSO such as basic structure, the basic theory, improvement and realization are systematically discussed in our work. The main contributions given in this dissertation are as follows:1. Due to excessive lose of PSO in evolutionary process with diversity of population and early premature convergence of the algorithm, a problem is caused about poor performance of global optimization in PSO, an improved PSO algorithm is proposed, where the dynamic learning factor and inertia weight based on particle's flying are combined to standard PSO, and fully reflect the idea of division of labor, and attain the aim to increase the diversity of population. Experiments upon extensive benchmark functions show that, the improved PSO is more efficient to solve the premature problem with higher global successive rate and accuracy.2. In PSO, the particles search the solution space to obtain the optimal solution with individual experience and group information. Because the obtained information is not sufficient and can not overcome the disadvantage of premature convergence of PSO, another improved PSO is introduced in the paper, which includes add-in that is added in speed formula to increasing the experience of particles, and adjusts dynamic inertia weight according to the speed of particles and aggregation degree of particles. Experiments indicate that this algorithm has faster convergence and higher computational precision.3. BP (Back Propagation) algorithm is a training algorithm in training artificial neural network. However, it is known to have some defects, such as complex gradient calculation, converging slowly and immersing in local vibration frequently. The thinking of the two improved PSO algorithms has been used to train artificial neural network in this paper. It has been proved by experiments that the two improved PSO algorithms have more classification effect than BP algorithm. In all, the paper makes an analysis of PSO algorithm in detail. It not only puts forward two effective modified measures but also broads its application areas. Finally, the paper summarizes the research work and proposes the directions for next research.