Research On Clustering Based On Artificial Immune Theory

The artificial immune system (AIS) is an emerging intelligent method and it's a calculation model designed on the basis of the biological immune system's functions, features and mechanisms. It has strong ability in study, recognization, memorization and character extraction, so it can solve many complex problems challenging the traditional calculation methods. Therefore, AIS has become a research focus in recent years and has been widely applied in many fields.In this paper, clustering problems are studied by the artificial immune system. The deficiencies of the exsiting clustering analysis algorithms are analyized and improvements are made accordingly. The main achievements of this paper can be summaried as follows:First, studies of evolutionary calculation in the aspect of clustering analysis are analyzed. The existing immune algorithm is improved and a new immune clustering algorithm based on the clone selection principle is presented. The algorithm adopts a mechanism that includes the differentiation of memory cells and suppressor cells and the production of suppressor antibodies by suppressor cells, preventing from copying of the memory cells. Meanwhile, the k-means algorithm is treated as a search operator in order to acquire the initial antibodies. Compared with k-means algorithm and the standard genetic algorithm, the clustering algorithm based on the clone selection principle has a faster convergence speed and can find out the global optimum solution.Second, typical immune networks are studied. Their deficiencies are improved and then a new immune network model called MDF-aiNet (a modified artificial immune network) is presented. The new model uses ART (Adaptive Resonance Theory Network) to acquire the initial antibodies for the network, and adopts a cross and mutation method more resembling the biological system's affinity maturation mechanism for solution research. Experimental results indicate that MDF-aiNet can compress data without damaging the original data structure and it has a higher compression ratio than the aiNet algorithm.Third, on the basis of MDF-aiNet, an immune network clustering algorithm is presented by combining FCM (fuzzy c-means) algorithm. And the simulation result indicates that the new algorithm has a higher clustering precision than FCM.