Research On PSO Algorithm Based On Self-adaptive Neighborhood Explored And Population Centroid Lenarning Mechanism

Particle swarm optimization is an optimization algorithm based on swarm intelligence presented by Kennedy and Eberhart in 1995. It simulates the action of the bird swram looking for food by flying and to get the optimization through the cooperation in the bird swarm. The algorithm obtains the evolution of all swarm from changing information between the individuals. The particle can adjust the moving track of themselves from the best postion itself ever went previously, the whole swarm ever went previously and the inertia term. Its major advantages are: simple concept, few parameters, supporting real-coded, requiring no gradient information, and easy to implement et al. So, it has a widely application.However, PSO also has its own shortcomings, the algorithm may fall into premature convergence and low accuracy, the main cause is every particle's speed can not be refreshed in the anaphase of optimization, which cause the particles collect together very tightly in some area and can not search more wide and meticulous. For the particle swarm, it can be said the swarm is be short of variety, little difference between each particle and less information canurges the swarm to develop, especially in high-dimensional modal optimization problem.There are two main characteristics about the improverment. First, most improvements are based on the core concepts of position and velocity, form different strategies to change or adjust the two variables. They may add some operators, which make the description more complex and the quantitative analysis of the convergence more complicated; second, in order to avoid falling into the local minimum, most improvements give a threshold, by determining whether the parameters reach it, using this method to change position or velocity. The adjustment is passive, it can not achieve the diversity, because most of the particles may near to the local optimum at the moment.Based on analysis the shortage of the PSO, it gives two improved strategies. Durning the particle evolution, to select some particles to detect the neighborhood of the global optimal postion by decreasing the probability, and the mutation of velocity mechanism also can be conducive to the evolution.Durning the particle evolution, only considering the best particle position of the swarm is inadequate, but also need to consider the evolution information of swarm, using the swarm information to guide the evolution. So, another population centroid is calculated, the appropriate centroid information can make the next evolution direction be more conducive to the evolution.After a series of experiments, it is proved that the algorithm based on the improvement in this thesis has the high quality on precision, stability and convergence rate. It also indicates that such improved evolutionary can greatly overcome the shortcoming of low efficiency in traditional PSO algorithms. At last, an improved PSO algorithm is applied to the evaluation of circularity error.