Application Research Of Cluster Analysis Based On Ant Colony Algorithm

Cluster analysis, as one of the important research branch of data mining, has been widely applied to pattern recognition, data analysis, as well as image processing and many other domains. As a new intelligent bionic algorithm, ant colony algorithm has showed good prospect in the cluster analysis because of its inherent parallelism, robustness and self-organizing. This thesis studies and analyzes ant colony algorithm deeply, and proposes the corresponding improved algorithms and validates the validity of the improved algorithms.This thesis contains the following aspects:1. Description of clustering analysis in brief. This thesis mainly introduces the definition of clustering analysis, inter-object similarity measurement method, common method of cluster analysis and the clustering results evaluation standard.2. Description of the ant colony algorithm in brief. This thesis introduces the mathematical model of basic ant colony algorithm and points out further research directions and improvement aspects after the analysis of advantage and disadvantage of foraging behavior based and clustering behavior based ant colony algorithms.3. Fusion of the ant colony algorithm and improved genetic algorithm. At first, an improved genetic algorithm, which has the characters of pseudo-parallel, populations dynamic adjusting and the optimal solution preserving, is proposed. Then fitness function and genetic operators fit ant colony clustering are designed. This thesis fuses ant colony algorithm and improved genetic algorithm, using genetic algorithm to determine the optimal parameters and initial pheromone distribution for ant colony algorithm, using ant colony algorithm obtain clustering results. The validity of the algorithm was validated.4. Research on performance improvement of ant colony algorithm. This thesis proposes a modified version, which improves the two-dimensional grid, short-term memory, ants pick up and place the objects strategy, parameters adaptive changing strategy and discrete objects dealing strategy. The validity of the algorithm was validated.