Optimization Algorithm On Artificial Neural Network Research And Application

As a new kind of Swarm Intelligence Algorithms, Particle SwarmOptimization(PSO) alogrithm is valued by scholars both at home and abroad for itssimple calculation,easy implementation, fast convergence speed and goodconvergence performance, and it is applied to function optimization, neural network,data mining, fuzzy system control, with good prospects on engineering.This paper focus on the deeper research about the improvement of ParticleSwarm Optimization algorithm and its application in artificial neural networktraining.The main contents include:1. In this paper, Particle Swarm Optimization algorithm s basic principle,algorithm process, social behavior analysis and two kinds of evolution model wereintroduced, four common improved Particle Swarm Optimization algorithm wassummarized, and its main faults were analyzed.2. In view of the shortcomings of Particle Swarm Optimization that it is easilyfalling into local extremums, the Harmony Search algorithm theimproved ParticleSwarm Optimization algorithm was proposed. Harmony Search algorithm is a newkind of optimization algorithm having good global searching performance. In thispaper, theparticle swarm of Particle Swarm Optimization algorithm was treated as theharmony memory of Harmony Search algorithm. The particle was convented to newone according to the harmony search algorithm, and the the worst particles in originalparticle swarm were replaced by comparing the fitness value, then the position andvelocity of particle were updated according to the original evolution equation ofParticle Swarm Optimization algorithm. Through the test of four standard testfunctions, the Particle Swarm Optimization algorithm that blending Harmony Searchalgorithm is obviously enhanced both in convergence speed and convergence precision.3. Particle Swarm Optimization algorithm was applied to the Neural Networktraining, and nonlinear function fitting was conducted to the trained Neural Network.Through Matlab simulation, the results showed that the effect of improved ParticleSwarm Optimization algorithm that blending Harmony Search algorithm for NeuralNetwork optimization is better than the basic Particle Swarm Optimization algorithm.In Neural Network training process, the improved Particle Swarm Optimizationalgorithm is obviously better than basic Particle Swarm Optimization algorithm onparticles convergence rate and robustness, however the convergence performace isslightly lower than basic Particle Swarm Optimization.