Genetic Algorithm-based Multi-objective Optimization Problem

Multi-objective optimization has been a difficult problem and focus for research in fields of science and engineering. There already have a lot of classical methods for solving multi-objective optimization problems before evolutionary algorithms were introduced in 1985. Classical multi-objective optimization methods have been thoroughly developed, but there are still lots of shortcomings in solving high dimension, multimodal problems. GAs can handle large space of problem and get a lot of trade-of fronts (possible solutions) in one evolution. A GA does not need much information about the problem before starting the optimization process, also it is not sensitive to the convex of the defined fields of the objective functions. So using GAs in solving multi-objective optimization problems is the most important research direction in the future.Based on extensive and deep review of literature, a thorough analysis and research on many theoretical and application oriented problems is presented. The main contents are as follows:Firstly, we introduce the multi-objective optimization theory, bring forward the model of multi-objective optimization and explain the concept of the pareto optimization solutions. We summarize some classical multi-objectives optimization methods and show the limits of them. In this paper, we also introduce the basic theory of genetic algorithm, including its working flow and the common technology that has been used in the optimization process. The schemate theorem and building block hypothesis are also expatiated in this paper.Secondly, We summarize some classical multi-objectives optimization methods based on genetic algorithm: VEGA, MOGA, NPGA, NSGA, NSGA-Ⅱ, SPEA and introduce the methods of allocating fitness and keeping the variety of population.Finally, we import knowledge of immune, co-evolution and game theory into genetic algorithm to improve the performance on solving the multi-objective optimization problems. The results of the experiments show that all of them can get better results than the original algorithm.