| Particle Swarm Optimization is an intelligent evolutionary computation algorithm widely in application .The algorithm achieves Optimization through paiticle tracking its best position and the best swarm position . Particle swarm optimization is simple,easily achieved and only need to adjust very a few parameters, it can obtain accurate result in a short time. Particle swarm optimization has become a popular issue in research work of optimized technique application domain in currently. As a late-model simulative evolutionary Algorithm, many key problems will be needed to study, for example: convergence speed, convergence time, earliness problem, theoretics bases and so on.The inertia weight is the most impordant adjusted parameter in particle swarm optimization. The size of inertia weight decides how many currently speed particle to inherit , bigger value will make partile having quick speed, it can improve the search ability of the algorithm wholly ; and that, small size of inertia weight will intensify enhancing the search ability of the algorithm locally, so it will make for convergent control. To a extraordinary extent, proper choice of inertia weight size determines the effect of algorithm execution. It is very necessarily to make a study on inertia weight.As a wholly efficient search method, particle swarm optimization has achieved good application effect on a lot of problem,such as nerve network training,robot routing programming, signal handling and mode identify, compounding optimization, multi-goals optimization, Automatic aim detect, biology signal identify, decision-making adjusting, system identif and so on. Lately, making particle swarm optimization applify to parameter estimate has been becoming a hotspot.Parameter estimate in Regression analyse is a case that random variable distributing function has been existed, but parameter is unknown in practical question. If a set of sample value has been obtained, we wish use it to estimate parameter value of the distributing variable. It is a very important problem on engineering. In regression analyzing , maximum likelihood estimation method is a basis way for model parameter estimating, but, when the method is used in parameter estimation, it is demanded to solve simultaneous equations in a general way, it is complex and difficult to solve by conventional iterative algorithm, while the convergence of algorithm is bad, even sometimes it can not be constringent.In this paper, on the basis of principles analying of PSO , contraposing standard PSO getting into earliness locally more easily, a improved PSO(Non-Linear Decreasing Random Inertia Weight PSO) are proposed based on modifying the inertia weight of standard PSO. It is a new strategy of inertia weight to add consider of random factor based on Non-Linear Decreasing Inertia Weight. Now the new algrithm is coming to appled in parameter estimation of multiple linear regression models and Logistic,Probit Non-Linear regression models. The effectivity and advantage of this algorithm is verified by numerical simulation computational experiment.The main research work and contribution are as follows in this paper:1. At present ,Several basis theoretics and improved methods of PSO have been introduced generally, based on it, the problem about earliness of PSO has been analysed .By studying the setting of the most impordant adjusted parameter (inertia weight), a new idea is put forward, an improved algorithm,called as NLDRWPSO, is proposed in which the capibility of particle's resisting logical optimization is greatly strengthened. Experiments on benchmark functions show that the performance of NLDRWPSO outperforms standard PSO.2. The improved PSO(NLDRWPSO) will be applied in parameter estimation of regression model, such as multiple linear regression models,Logistic, Probit Non-Linear regression models.An evaluation report will be given about performance of new algorithm by comparing other intelligent method or traditionary mathematic ways through practical examples. |
PDF Full Text Request