Improvement Of Real-coding Genetic Algorithm And Its Applications

Since the Genetic Algorithm was proposed by Professor Holland at MichiganUniversity,its theories and applications have attracted attention of many scholars.Function optimization is one of the classical applications of the Genetic Algorithm.However, there are some disadvantages when applying Genetic Algorithm to globaloptimization, such as premature convergence, large amount of calculation, low accuracyon solution and low convergence speed, specifically when solving problems aboutmulti-peak and multi-dimensional complex functions. As a result, many scholars havemade efforts to improve the Simple GeneticAlgorithm, in which situation improvementon the genetic operator is most commonly studied. This paper mainly concerns aboutthe improvement on real-coding Genetic Algorithm and its applications to functionoptimization.First, we introduce the mathematical knowledge of Genetic Algorithm. Second, wedescribe some recent advances on the Genetic Algorithm and briefly introduce somenew improvements on the Simple Genetic Algorithm done by other scholars. Then, weanalyze the advantages and disadvantages of these improvements.Due to low and mutations of the optimization deficiency in solving multi-peak andmulti-dimensional functions and global optimization problems, a modified geneticoperators is proposed:(1) Crossover operators:This article analyzes the advantages and disadvantages ofthese improvements proposed by the precious scholars on the base of some newcrossover operators. The improved crossover operator not only is as effective as thetraditional crossover operator, but also increases the diversity of the population. So,theimproved crossover operator plays a role of mutation operator.(2) Mutation operator: This article made some improvements on mutation operatorbased on the adaptive change of mutation proposed by bibliography[53].The improvedmutation operator is easy to operate and can improve the efficiency of GeneticAlgorithm.(3) If improperly selected, fitness function throughout the optimization process ofGenetic Algorithm will lead to deceptive problem. This article proposes a new fitnessfunction based on the smooth function and exponential transformation, which bringsabout better results and more versatile. (4) In order to test the effectiveness of the modified Algorithm,we select sixdifficult multi-peak morbid functions and six multi-dimensional functions. Wecarry them out in the Matlab7.1environment and compare them with other GeneticAlgorithm. The results show that the improved Genetic Algorithm performs better.