Design and application of wavelet networks based on immune algorithm

Based on the analysis of evolutionary algorithm´s properties, a novel learning algorithm of wavelet networks with the concept and theory of biological immune system, namely immune algorithm (IA), is presented. In order to keep the diversity of antibodies, and improve the global search ability and convergence speed, IA stimulates or restrains antibodies according to the density and affinity of the antibodies in the pool. An IA based wavelet network is applied to continuous function approximation, and the computational process is compared with that of evolutionary algorithm and BP algorithm. The results show that the immune algorithm can greatly improve the convergence and achieve higher accuracy.