Multi-objective Clustering And Applications Based On PSO And Immune Optimization

With the rapid development of information technology and networks, data generated in people's daily lives has increased in the form of positive exponential. How to explore useful information from these massive data becomes one of the hot topics in recent years.Clustering has been widely concerned as a strong force tool of data analysis. A variety of clustering algorithms are emerging. This thesis takes multi-objective clustering as the overall framework, and the study of data clustering, face clustering and image segmentation method is made by means of particle swarm optimization and artificial immune method. The main contents include:1. A multi-objective immune particle swarm optimization clustering algorithm with particle regeneration strategy based on age is proposed. The new algorithm combines particle swarm optimization with immune optimization and optimizes between-cluster separation and within-cluster compactness simultaneously. In the process of particle swarm optimization, we propose a new concept of age to measure the local search capability of each particle. Meanwhile, to avoid the new algorithm falling into local optimum,a new particle regeneration strategy based on age is proposed. In the process of immune optimization, we utilize clone operator, crossover operator and mutation operator to increase the diversity of population, which effectively avoids the prematurity of the algorithm.2. A multi-objective artificial immune algorithm for fuzzy clustering based on multiple kernels is presented. The new algorithm unifies multi-kernel learning and multi-objective optimization in a joint clustering framework, which preserves the geometric information of the dataset. This approach is effective, not only for spherical clusters, but can also discover the non-linear relationships between data, and adds robustness to the particular choice of kernel functions. Additionally, the introduction of multi-objective optimization helps the proposed algorithm to avoid becoming stuck at local optima. The experience results on UCI datasets and face datasets suggest that the proposed algorithm has a wider scope of application.3. A multi-objective artificial immune automatic clustering algorithm based on dynamic local search is proposed. Multi-objective optimization is introduced into the new algorithm.And it optimized the number of clusters dynamically. In the clustering optimization process, we utilize artificial immune algorithm to solve the multi-objective automatic clustering problem. It can avoid the phenomenon of degeneration and premature of traditional evolutionary algorithms and improve the ability of the algorithm to search the global optimal solution simultaneously. Meanwhile, we also introduce a dynamic local search operator, so that the number of clusters can update dynamically. The dynamic local search operator speeds up the convergence and improves the clustering accuracy of the proposed algorithm. Experimental results on artificial datasets, UCI datasets and synthetic images show that the proposed algorithm is more suitable for automatic clustering problem compared with compared algorithms.