Chaotic Coding Particle Swarm Neural Network Research

Particle swarm optimization (PSO), with straightforward, few parameters, quicker convergence and easy to be realized, gets a really nice at lots of scholars. Particle swarm optimization is a new globe optimization algorithm based on swarm intelligent search. But it is easy to fall into local minimums. Particle swarm clump together and get stagnation, which caused algorithm trap in premature convergence. Several kinds of improved methods were proposed in this paper, and the chaos number was also be used in particle swarm optimization, and at last, the improved algorithm was used in neural network training to solve classification problem.In the particle swarm optimization algorithm, particles velocity is updated just by its current velocity, the best position in its experience, and the best position in the swarm. Considering the previous changes influence among the current particles velocity, in this paper, a new particle swarm optimization algorithm with momentum factor was proposed. In the new algorithm, a momentum factor was introduced to the particle velocity changing formula. The current particles velocity changes are influenced by the previous changes. Let the algorithm have a quicker convergence velocity.In the basic particle swarm optimization algorithm, particle position update can be considered as flight distance per unit time. In the different time, the particle position update distance is different, and its affection for algorithm is also different. A time parameter is introduced in this paper, in the initial stage, increase the particle position update every time, and decrease in later stage. The simulation result shows that the time parameter improved the convergence rate of the algorithm.PSO algorithm has good search ability on the global solution and converges fast at the previous period, but it is unsatisfactory in the later stage of searching optimal solution. BP algorithm has a great ability of local searching. Scale factor λ1,λ2to assign the different weight to PSO and BP algorithm during different periods are proposed in this paper. And the sacle factor λ1, λ2are tuned through fuzzy control. When the network error is big, the degree of convergence of the network is small, then the PSO algorithm plays a main role. When the network error is small, the degree of convergence of the network is big, then the BP algorithm plays a main role.As a new arithmetic coder, chaos number can describe object developing process accurate via its dynamic implication factor. In this paper, chaos number is used in particle swarm algorithm. Chaos number can describe birds characteristics better, and make the algorithms is closer to birds activities, and can imitate birds intelligent more accuracy. At last, use the chaos number algorithm training the neural network, and resolve the classify problem, the result shows that it is better than real number.