Particle Swarm Hierarchical Clustering Of Different Groups Of The Number Of Particles Were Analyzed

This thesis focused on discussing the research situation of domestic and foreign scholars for particle swarm optimization algorithm and its shortages, by which introduced the particle swarm optimization algorithm based on hierarchical clustering thought, and improved the algorithm. The algorithm clustered the different levels of groups, and put the best individual in each groups to the next layer. The improved algorithm compared the best individual of each groups with their correspondent individual in next layer to substitution over if better, which made the next layer groups have a better optimization. Meanwhile, the algorithm introduced the mutation operator and deceleration factor. The algorithm can increase the diversity of the population particle, and avoid the local optimum in the search process, and faster searching the global optimum position through iterative.By comparison and analysis of the paper, it used14typical experimental test functions. In the experimental design, it analyzed the characteristics of each test function to divide the14test functions into7types test environments in the three different spatial dimensions which are2-dimensional,10-dimensional and100-dimensional. For seven different test environments on the basis of the improved algorithm, the paper made comparative analysis of the inter-experimental for the size of each groups of particles, to discuss the relationship between the size of each groups particles and particle swarm optimization algorithm ability to search for the optimal value in different environment. And finally, it summed up the three general conclusions:1) In the lower dimension, particle swarm optimization algorithm ability to search for the optimal value would be better in the less number of particles of each group.2) In the dimension environment between low and high-dimensional, particle swarm optimization algorithm ability to search for the optimal value would be relatively good both in the less and more number of particles of each groups. And the search ability would be relatively weak in this between the two cases. 3) In the lower dimension, particle swarm optimization algorithm ability to search for the optimal value in each group would be better in less number of particles.At the end of the article, the paper applied the conclusion to the high dimensionality of gene classification of tumor. Through the simulation experiments, it more reflected the superiority of the improved hierarchical clustering of particle swarm optimization, and the accuracy and stability of the expression of genetic data classification process.