Distributed Imniune Evolutionary Algorithm For Function Optimization Problems

Many problems encountered in science and engineering applications can bemodeled and treated as function optimization problems(FOPs).So study on functionoptimization problems is very important in theoretic and practice. Immune Algorithms(IAs) are a kind of simulated nature algorithms, which developed based on ArtificialImmune Systems-a kind of computation intelligence system occurred in recent years.Evolutionary algorithms(EAs) is a kind of randomly search algorithm which developedby simulate natural evolutionary mechanisms. In recent years, this two typesof algorithms have been widely applied to solve function optimization problemsand generated a lot of function optimization method.Firstly, this dissertation introduces the general theory and its application ofartificial immune algorithm and evolutionary algorithm. Then an summary aboutadvantages and disadvantages of the two kinds of algorithm while being applied infunction optimization problems. The feasibility of integration of advantages of the twotypes of algorithm for solving high dimensional function optimization problems isanalyzed.Secondly, this dissertation designs a new model—new master-slave model(NMSM)on the research of memory mechanism of immune algorithm and the distributed modelof evolutionary algorithms. In this model new strategy of elitist-selection andelitist-crossover is used. The slave population and the main population can executeevolutionary process in parallel. And a new algorithm called distribute immuneevolutionary algorithm(DIEA), which includes one the master module and several slavemodules, is designed based on the NMSM. The master module is responsible for theevolutionary process of elitist antibody, while the slave module is responsible for theevolutionary process of sub-population. The improved operators of immune algorithmand evolutionary algorithm are used in this module. Then the overall performance of thealgorithm is analyzed and the convergence of DIEA is proved.Finally, the dissertation designs a kind of multi-threaded simulative parallelcomputing system and verified the solution quality and convergence speed of DIEAthrough the different test function. Finally, we give a conclusion of the researchwork and give some advice for further improvement and application.