Research On Improved Particle Swarm Optimization And It Used In Artifical Neural Networks

Particle Swarm Optimization is an new intelligence algorithm base on thesimulation of biologic swarm social activity.The algorithm is the random and uncertain,so,the theory of algorithm is not complete. It also has many defects,such as easying topremature at the initial iteration, slow convergence speed at the later iteration. The searchon analysising theory and improving algorithm has become to the most important andhotpots.In this paper,combined with the biological prototype characteristics,and based onthe triditional particle swarm optimization algorithm,the author introduced two improvedalgorithm for the convergence performance.The triditional Back Propagation learningalgorithm is one of artifical neural networks,it is an optimization algorithm based ongradient information.But it was easy to falling into the local optimal.The improved particlesawarm algorithm was applied into neural network training.In this paper,it was not onlyimproving the theory of PSO algorithm,but also providing the related works in the PSOalgorithm applied into Artificial Neural Network.The research work of this paper was asfollows:1. The summary of optimization problem.It is basing on mathematical theory, forsolving the optimal solution in various practical applications. Optimization problems andalgorithms are the theoretical subject and the application value subject.the traditionaloptimization algorithms has Lagrange multipliers,the complex method, conjugate gradientmethod and so on.But in practical applications,it often has large-scale, high difficulty andmany local optimal characteristic.Therefore,the traditional algorithm is difficult to resolvethis problem.Intelligent optimization algorithm is an method,by simulating a biologicalsystem and its behavior,it is an strong generic optimization model and can solve somecomplex optimization problem. PSO is one of intelligent optimization algorithms, it is aheuristic stochastic optimization algorithm, using cooperation mechanism between thegroup for iterating to produce the optimal solution.The PSO algorithm is simple concept,easy to implement, few parameters to adjust,and it is to become a hot topic at home andabroad, especially in the theory analysis, improved algorithm and its applications.2.PSO is a stochastic,with heuristic information optimization algorithm. Theinteraction between the group of particles, the particle trajectory is very complicated. From a theoretical point of view, if the random algorithm can convergence for probability1,thisis a satisfactory result. In this paper,Analysis of the influence of parameters on theconvergence, convergence speed and convergence accuracy. Analysis on PSO behavior andgive the convergence region parameter.Providing the parameters choose criteria for thelater research,at last proving the convergence position of the algorithm is existence anduniqueness.3.PSO is easy to premature convergence.The reason of slow convergence rate at thelater is indicidual in groups update extreme slowly,and lack of vitality.So,For improvingthe performance of algorithm,An improved constriction factor particle swarm optimizationalgorithm was proposed to overcome the local optimum in this paper.The improvedalgorithm has introduce two new parameter, a position factor and speed factor.If allparticles are close to the global optimum and its velocity is less than the speed factor that itis set in advance,these particles are likely to be stagnate,then a reinitialization to enhancethese particles’ energy.When the algorithm fall into local optimum,this method candisperse the particles and improve the diversity of the population and avoid prematureconvergence.The three Multimodal functions’ simulation, the data can verify theperformance of the improved algorithm, and shows that the improved algorithm improvesits convergence accuracy, even, can show that the improved algorithm can effectivelyavoid falling into local optimum.4.The bacteria have many features,such as short life cycle, reproduce rapidly andsensitive to environment.After foraging sufficient nutrients,bacteria will be the seconddivision multiply,but if the bacteria do not feed enough nutrients,bacteria can easily died.Anovel version of PSO algorithm, called bacteria PSO (BacPSO), was proposed in this paper.In the new algorithm, the individual was replaced by bacteria, and a new evolutionarymechanism was designed by the basic law of evolution of bacterial colony.Suchevolutionary mechanism also generated a new natural termination criterion. Propagationand death operators were used to keep the population diversity of BacPSO.The simulationresults show that BacPSO algorithm not only increasing the convergence speed,but alsocan converge to the global optimum. 5.The Artificial Neural Network is one of the important systems of artificialintelligence research,it has a parallel to the height of the graph topology, dynamicalsystems and exert the role of the input state to obtain the output.The Back Propagationalgorithm (BP algorithm) is one of the most important, the most widely used neuralnetwork algorithm, which uses optimization techniques is the most common gradientdescent.The basic idea of least squares learning algorithm is based on the gradient searchfor the minimum variance algorithm,in order to obtain the minimum mean square errorbetween the actual network output and the desired output, the network learning process isthe error from the upper to the lower again spread again the value of the right to amend theprocess.However,in many practical problems,the complex nonlinear gradient informationof the object is difficult to get unable to obtain,so the BP learning algorithm can not beoptimized to solve the problem.This article will improve particle swarm optimizationapplied to the BP learning algorithm, the combination of the characteristics of the BPalgorithm and PSO algorithm global search capability.PSO algorithm instead of the initialoptimization of the neural network,the network is only close to the optimal solution have toparameter optimization based on simulation results,the PSO based neural networkalgorithm significantly improved network optimization accuracy and speed.