Application Of Group Intelligence Optimization Algorithm In Cluster Analysis

Swarm intelligence optimization algorithms belong to the intelligent calculation methods by simulating the behavior of biological swarm intelligence, which have good optimization abilities, search capabilities, and easy to implement, which are flexible and practical. The efficiency is better than the traditional calculation methods in solving large-scale complex problems and optimization problems. So they have been paid much attention by the researchers and become an advanced research hotspot. More than ten swarm intelligence optimization algorithms have been proposed during the 2 decades from the first swarm intelligence optimization algorithm proposed, they have been successfully applied to function optimization, project scheduling, image processing, traveling salesman problem, and logistics location, clustering analysis, bioinformatics research and other fields.Clustering analysis is one of the basic methods of understanding and analysis the data, finding relationships and the inherent laws among data. In recent years, it has been extensively applied to the researches and applications in machine learning, data mining, pattern recognition, image processing, biology, statistics, market analysis and many other fields, and plays a very important role in promoting the development of science, technology and social progress.This paper gives a brief overview of the advantages of swarm intelligence optimization algorithm, introduces the bionic principle and algorithm process of four kinds of swarm intelligence optimization algorithms in detail such as ant colony optimization algorithm, particle swarm optimization algorithm, fire fly optimization algorithm, cuckoo search algorithm. Then the requirements and the classification of clustering analysis algorithm of the traditional clustering algorithm were discussed. On this basis, establishes the corresponding clustering analysis model and the corresponding clustering analysis algorithms based on swarm intelligence, applies these algorithms to clustering analysis to improve the shortcomings and proposes improved algorithms, simulation experiment and study are carried out. The following three parts of the work have been completed in this thesis:In the first part of the research work, designed a simple logistics distribution center location model, designed clustering algorithm based on ant colony algorithm and applied it to the logistics distribution center location model.Firstly, the clustering algorithm based on ant colony algorithm is designed for solving logistics distribution center location problem. Logistics distribution center location problem belongs to the clustering problem with high application and practicability, on the basis of referring to the books and articles, simple logistics distribution center location model is designed, with the combination of the ant colony algorithm and logistics distribution center location model, the corresponding clustering algorithm based on ant colony algorithm is proposed to address the logistics distribution center location model. The simulation experiment results are compared with K-means algorithm. This part of work has laid a good foundation for further studies on clustering analysis using swarm intelligence optimization algorithm.In the second part, we designed a clustering algorithm based on the firefly algorithm and the simulation experiments on UCI data sets, but the results showed that the convergent performance of the algorithm is not very stable, and the convergent rate is slow, easy to fall into local optimal value, therefore an improved firefly clustering algorithm is put forward based on optimal class-center disturbance. The movement and randomization of the firefly is modified, and then the values of step factor are compared in the experiment.In the third part, Levy flights are adopted to the cuckoo search algorithm, which can greatly enhance the search ability of the algorithm. But the convergence and stability are not ideal when we cluster using cuckoo search algorithm, in order to solve this problem, the idea of "social learning" of PSO algorithm is introduced to the cuckoo search clustering algorithm, a new algorithm for clustering is proposed. Then testified the performance by the simulation experiments on UCI data sets, the experimental results show that the new algorithm has strong global optimization ability like cuckoo search algorithm, but also has stable convergence ability like particle swarm optimization algorithm, both search capability and convergence performance have been significantly improved.