Research On Hybrid Improvement Of Genetic Algorithm And Its Application

Genetic algorithm is a probability search method which simulates the process of natural selection and biological genetic evolution.It has characteristics such as strong robustness, good global search performance,easily realizing parallelization. After years of research and continuous improvement,genetic algorithm has possessed a common framework for solving combinatorial optimization problems and it has been widely applied in production control, pattern recognition, artificial life, machine learning and other fields,it has gained lots of achievement.However,the inherent disadvantages of genetic algorithm are gradually exposed in the process of application, it is mainly reflected in three aspects including easily falling into local optimal, poor local searching and fine tuning ability as well as harply narrowing of the differences between individuals in the middle and late of evolution.For years, domestic and foreign scholars have proposed improved methods for genetic algorithms from many aspects and achieved remarkable achievement, hybrid genetic algorithm and multi-population genetic algorithm are the focus and difficult areas of current research on genetic algorithm.Inspired by the achievement of previous research,this paper combines characteristics of hybrid genetic algorithm and multi-population genetic algorithm and put forward feasible and effective hybrid improvement strategy.The main content and work of this paper includes the following three parts:(1)As to easily falling into local optimum, poor global optimization,excessive loss of individual diversity of genetic algorithm,a multi-population DNA annealing genetic algorithm(MADNAGA) is proposed.Character set which consists of four DNA base is adopted to encode the parameters of problem to be solved,based on this way of encoding,crossover,mutation and inversion operator are designed.Three population coevolution structure which is made up of two development population and an elite population is adopted and different genetic evolutionary mechanism as well as evolutionary control parameters are designed for them,information exchange among populations is realized by population interactive strategy. In addition, simulated annealing mechanism is introduced to realize the replacement of individuals.Finally,eight function optimization experiments and the comparison of optimization results are adopted to verify the effectiveness and superiority of the new algorithm.(2) With DNA encoding and three population coevolution structure,way of communication of populations as well as the generation and selection of individual are further improved, a muti-population immune DNA genetic algorithm (MIDNAGA) is proposed. Biological immune mechanisms is introduced for the adaptive regulation of the process of generation and selection of individuals. Finally, the validity and superiority of the new algorithm is verified by function optimization and the comparison of optimization results.(3) The two proposed improved hybrid genetic algorithm in this paper are applied to solve the knapsack problem, greedy criterion is introduced to repair and optimize the genes of individuals generated during the process of genetic evolution, simulation results indicates that the new algorithm show effectiveness,practicality and superiority on solving knapsack problem.