Biological Immune Based Immune Optimization Algorithms In Dynamic Environments And Their Applications

While intelligent algorithms are widely applied to static optimization problems, a considerable number of valuable achievements have been reported. However, It is studied rarely to deal with dynamic optimization problems, in which the key solving this class of problems is to design intelligent optimization techniques capable of strongly tracking the changing environment over time and achieving reasonable trade-off between performance effect and efficiency. So, it has become an important research topic to design more advanced intelligent optimization techniques to cope with dynamic optimization problems. From the angle of intelligent optimization, recently, the main research work has been focused on modifying classical genetic algorithms, but less progress. Therefore, in this dissertation, based on the theory of biological immune systems, three kinds of immune optimization algorithms in dynamic environments are proposed for dynamic single-objective optimization, dynamic multi-objective optimization and online greenhouse control, respectively. These algorithms are examined through numerical experiments, comparative analysis and applications. The main work is summed up as follows:A. A novel immune optimization algorithm in dynamic environments is proposed to deal with dynamic single-objective optimization problems. In design of the algorithm, several operators are established, i.e., dynamic evolution relying on antibody learning, antibody rearrangement depending on gene drift, dynamical memory pool composed of many memory subsets built upon the immune memory characteristics and the function of dynamic maintenance in which the pool utilizes the average linkage to keep those excellent memory cells, and environmental identifier and generation rule of initial antibody populations related to dynamic surveillance. The algorithm possesses such properties as structural simplicity, feasibility, and dynamical regulation of the execution time for different environments. Experimental results and comparison illustrate its superiority including the effective trade-off between performance effect and efficiency as well as the potential for complex dynamical high-dimensional optimization problems.B. An online greenhouse control immune optimization algorithm is proposed to solve a class of classical greenhouse control problems with dynamic environments. In the algorithm, dynamic memory pool is designed to preserve the excellent antibodies from the previous environments by using the dynamic update mechanism of memory cells in the immune system for reference, while the sizes of evolving populations are adjusted and their antibodies are chose dynamically in terms of the average densities of the populations. Besides, antibodies propagate their clones by means of their affinities associated to the given antigen. Through comparison with several evolution algorithms in dynamic environments, numerical experiments show that the proposed algorithm can track strongly changing environments with great practical perspective.C. A dynamic multi-objective immune optimization algorithm is proposed based on the characteristics of dynamic multi-objective optimization and associated to some metaphors of the immune system. In design of the algorithm, some antibodies are chose to participate in evolution through sorting level selection, while the affinity of an antibody is proportional to the average density of all antibodies in itsζ-neighborhood, being dependent on the position of the antibody. On the other hand, each clone undergoes mutation with its mutation probability conversely proportional to the affinity of its parent, and such immune functions as immune memory and dynamic maintenance, together with the average linkage method, are used to design environmental memory set and memory pool. Through comparison with two representative evolutionary algorithms and a neighbor search algorithm, numerical experiments illustrate that the algorithm perform well over the compared algorithms with respect to the speedup of tracking dynamic environments and its convergence.