The last step in training phase is refinement of the clusters found above. Although DynamicClustering counters all the basic k-means disadvantages, setting the intra-cluster similarity r may require experimentation. Also, a cluster may have a

多目标粒子群算法程序，感兴趣的可以进去-Multi-objective particle swarm optimization procedures can be of interest to go look at the

Multiobjective Particle Swarm with Crowding Distance（多目标优化）-Multiobjective Particle Swarm with Crowding Distance (multi-objective optimization)

Adaptive Multi-Objective Particle Swarm Optimizer