Artificial Immune System With Its Applications In Function Optimization And Data Clustering

Biological immune system is a highly intelligent distributed cooperative and adaptive system. It is a hot topic now to develop an intelligent algorithm based on biological immune system as well as its application in practical engineering backgrounds. In this thesis, the theory of artificial immune system as well as its applications in multi-modal optimization and data-mining is studied in depth. Inspired by the principle of immune system's antibody concentration suppress, the clonal selection algotithm (CLONALG) formulated by de Castro is modified accordingly. After the introduction of antibody's concentration, the variety of the network's population can be indexed thus the mutation rate of clonal selection algorithm can be made adaptive. Compared with CLONALG and opt-aiNET, this improved clonal selection algorithm maintains the optimization speed and high global optimization capability on one hand; it can also keep the diversity effectively on the other hand. The result is domenstrated in the simulation of optimizing some multi-modal functions.Furthermore, this thesis gives a detailed study of de Castro's artificial immune network's (aiNET) clustering algorithm. Since this algorithm is a very powerful tool in data compression and clustering, it is used here to get the training samples of T-S fuzzy-neural network clustered. It is this clustering process that determines the architecture of the network concerned. Then least-square method and back-propagation algorithm are employed to identify the coefficients undetermined. This is the procedure of fuzzy-rule extraction. To show the effectiveness of this algorithm, three typical benchmark problems are studied and simulated here. The results verified that this method is valid. Thus this algorithm establishes a new alternative to design fuzzy systems when the initial empirical knowledge is absent which usually leads to the difficulty in fuzzy rules extraction.