Multi-objective Optimization Immune Algorithm And Its Application In The Dynamic Environment

Biological immune system is a highly parallel adaptive information processing system that can identify adaptively and remove the intruding antigenic materials.This system has capabilities of learning,memory,adaptiveness and maintaining stability inside the body.Recently,many researchers begin to realize gradually the importance of the biological immune system to exploit intelligent algorithms,while it has become a research focus how to explore new optimization algorithms based on the immune system.Under this background,this dissertation proposes several immune algorithms to deal with different types of multi-objective optimization problems in dynamic environments.Through numerical experiments and practical applications,the results illustrate that these algorithms are available and effective.The achievements of this dissertation acquired are summed up below.Ⅰ.For a class of dynamic non-constrained multi-objective optimization problems with time-varying dimensions of decision space,a dynamic multi-objective optimization immune algorithm is presented based on the mechanisms of dynamic evolution of immune response.In the design of the algorithm,several crucial adaptive immune operators,such as antibody affinity evaluation,environmental recognition rule,affinity mutation,immune selection and environmental memory update and so on,are designed to evolve the current evolving population. In addition,depending upon the three performance indexes proposed in this work,numerical experiments show that the proposed algorithm can obtain satisfactory performance.Ⅱ.A binary-encoded dynamic constrained multi-objective optimization immune algorithm is proposed to deal with a class of dynamic constrained multi-objective optimization problems with variable dimensions of design space.In the design of the algorithm,an environmental recognition rule is developed to step up the process of optimization for similar environments in terms of the function of antibody recognition,while the concept of constrained-domination is used to design the scheme of antibody evaluation and the operator of immune selection.Besides,mutation strategies based on two-level probability control are developed to mutate the appearing antibodies. Comparatively numerical experiments show that the proposed algorithm,compared to any of two popular algorithms,is of more satisfactory search effect and stronger capability of environmental tracking.Ⅲ.A real-encoded dynamic constrained multi-objective optimization immune algorithm is proposed to solve the above constrainted optimization problems.In the design of the algorithm, relying upon the functions of antibody recognition,antibody learning,memory,dynamic balance maintenance of the immune system and so forth,several adaptive immune operators are designed to cope with these optimization problems.Especially,the concept of non-dominance is used to design an antibody evaluation scheme to evaluate importance of antibodies in their evolving population.Depending upon the three performance indexes proposed above and other three popular algorithms compared,comparatively numerical experiments show that the proposed algorithm is of great superiority over the three algorithms with respect to its performance effect and capability of environmental tracking.