Particle Swarm Optimizer And Its Application In Artificial Neural Network

Particle Swarm Optimizer (PSO) is an adaptive stochastic optimizer based on swarm searching strategy and was originally introduced by Kennedy and Eberhart in 1995. As a typical representative of swarm intelligence algorithm, PSO has been proven to be a powerful global optimization method. Now, it has been widely applied in Function Optimization, Artificial Neural Network, Fuzzy System Control and Pattern Recognition etc. The PSO algorithm has a good future in engineering application.This thesis mainly focuses on the improvement of PSO algorithm and its application in Artificial Neural Network, the main content of the thesis includes:1. Introduces the principle of PSO and improvement approaches, and concludes four approaches which can improve the algorithm performance, and every approach is introduced in detail. Then presents the experiment analysis and mathematical analysis of the convergence behavior of the PSO, which explains the reason of converging too early;2. To overcome the stagnation phenomenon in Cooperative Particle Swarm Optimizer, an improved Cooperative Particle Swarm Optimizer, Cooperative Particle Swarm Optimizer Based on Particle's Spatial Extension (SE-CPSO) was presented. In SE-CPSO, radius of particle and concept of jumping factor are introduced. If two particles collide, a strategy of jumping out of position is introduced to avoid the stagnation behavior. The results of testing on multi-functions show: the best parameter compounding of this algorithm are different for different functions, but all can effectively conquer the stagnation, enhance the ability of searching better solutions, and to a great extent can improve the robustness of algorithm and searching efficiency. And then, compares SE-CPSO with GCPSO-CPSO algorithm on some aspects such as convergence rate, robustness and so on, the result shows that the running velocity of SE-CPSO is quicker obviously, and the robustness is stronger;3. Applies Cooperative Particle Swarm Optimizer to the training of Neural Network, there are mainly two kind of Neural Network, Summation-unit Neural Network and Product-unit Network, and tests the performance using Iris standard classify data set, the result shows that Cooperative Particle Swarm Optimizer can train Neural Network as well as other technologies, and at the same time shows in two kinds of Neural Network, the best split factor of which Particle Swarm Optimizer trains Product-unit Network is smaller than that of Summation-unit Network.