Study On Immune Algorithm And Its Application

Immune algorithm (IA) was an new technology of information processing, which extracted from the characteristics of biological immune system, such as self-adaptation, self-organization, diversity, immune memory, etc. It was a hotspot after article neural network(ANN)and evolve calculation(EC)in the field of calculated intelligence. In this dissertation, the theory of IA and its applications in engineering problems were studied. The main contributions of the dissertation were as follows: In chapter 1, the origin and development of IA were overviewed. The biological background of IA was introduced. Its characteristics, applications and research status quo were generalized. In the end, the meaning and mainly content in this thesis were discussed. In chapter 2, aiming at complex calculation process and low searching efficiency of standard immune algorithm based on entropy, we present an improved immune algorithm (IIA) by directly calculating the vector distance between antibodies and fully selecting the next populations based on the density of antibody, which had the characteristics of concise calculation process, flexible coding, and no need linear change. In chapter 3, the mathematic model of IIA was put forward, which separating the process of state transfer of IIA into B-cell state transfer and immune memory cell state transfer. It was proved by means of homogeneous finite Markov chain that IIA had ergodicity property on B-cell state space and convergence on immune memory cell state space. In chapter 4, the advantage of IIA in the field of optimization was proved. Experimental results by comparing the optimization of a set of complex, constrained, nonlinear, indefinite numerical functions with standard genetic algorithm (GA) and simulated annealing (SA) showed that the IIA not only could effectively preserve diversity than GA in population but also had faster speed than SA in convergence. In chapter 5, an optimizing method searching the satisfied weights of the feed-forward neural network based on IIA called immune neural network (INN) was presented, which used the immune reproduction operation to maintain the diversity of solutions and utilized a new crossover strategy to minimize the location of progeny and embedded the traditional back propagation (BP) algorithm as a mutation operation to expedite convergence. Simulation results illustrated that INN has better capability of escaping from local optima and faster speed in convergence than evolutionary neural network (ENN).