Research Of Search Strategy In Immune Memetic Algorithm For Multi-objective Optimization

In the fields of scientific research and engineering applications, the optimization problem is one of the most important topics. The Problem, that has only one objective function, is known as single-objective optimization problem. However, in many real world applications, the optimization problems always have multiple conflicting objective functions, that need to be optimized in the same time. So, we must take into account all sub-objectives to achieve the best trade off.Artificial immune system, which is a type of intelligent method, provides a novel approach for solving the multi-objective optimization problems. The applications of artificial immune system on the multi-objective optimization problems have attracted lots of the attentions of researchers. Multi-objective Immune Algorithm with Non-dominated Neighbor-based Selection (NNIA), which uses neighborhood-based non-dominated individual selection method, is an outstanding immune multi-objective optimization algorithm. In order to increase the probability of searching sparse regions of current pareto frontier, NNIA only selects a few relatively isolated non-dominant individual as the active antibodies, and then applies the proportion clone according to the crowding distance of active antibodies. But, there are two weak points in NNIA. One is that NNIA just apply re-combination operator and hyper-variation operator to generate new solutions, the search strategy maybe reduce the convergence rate. The other is that it can’t guarantee to obtain a high quality solution.In this thesis we designed some local search operators according to the distribution characteristics of the population during the evolutionary process. And then, we combine the local search operators and NNIA’s main frame to construct new memetic algorithms. The research works in this thesis are as follows.(1) An improved multi-objective immune algorithm, Memetic Immune Algorithm for Multi-objective Optimization(MIAMO), has been proposed. The pareto dominance-based local search operator, works in the early stage of evolution, can speed up the convergence rate, and it also can make the population to move towards the pareto optimal solution. Neighborhood differential operator, works in the late stage of evolution, can increase the diversity of the population, and it also can make the population to move along the pareto optimal solution.(2) Proposed another improved multi-objective immune algorithm, An Immune Clonal Multi-objective Optimization with Neural Network-based Convergence Acceleration Operator (MIACA). By artificial neural network, the algorithm maps the new solutions, generated by local search on the object space, from the object space to decision space. This modeling prediction method can reduce the function evaluation, and thus accelerate the evolution of population.Finally, our improved algorithms were compared with other outstanding multi-objective optimizations on some classical multi-objective test problems. Experimental results show that the performances of the improved algorithms are better than NNIA.