The Improved Genetic Algorithm Based On Cloud Model Research

Cloud Model is a transformation model between the qualitative and quantitative whichbases on traditional fuzzy set theory and probability and statistics, it combines the stochasticand the fuzzy properties, reveals the relevance of them, and is used to describe the concept ofnatural language in uncertainty. Genetic Algorithm is a random search algorithm which baseson the survival of the fittest and genetic mechanism, and its main features are high parallelism,independent on gradient information, group search strategy and so on. But the prematureconvergence is a problem that would affect the optimization effect of genetic algorithm,extremely when all individuals in the group converge to the same local optimal solution andstop evolving or the individuals closed to the global optimal solution are always eliminated,the algorithm cannot converge to the global optimal solution. The generating factors ofpremature convergence could be imported in every link of the process of evolution of thegenetic algorithm. In this paper, by using the cloud model with the randomness and fuzziness,several aspects of genetic algorithm is improved in order to overcome the prematureconvergence phenomenon:(1) This paper presents a multi-population adaptive genetic algorithm based on cloudcontrol. The setting of the crossover and mutation rate is one of the key problems in geneticalgorithm. This paper balances the two aspects of overall property and individual differences,realizes the adaptive adjustment of the crossover and mutation rates through the cloudcontroller. The experimental results show that, compared with the standard genetic algorithmand adaptive genetic algorithm, this algorithm can effectively avoid the prematureconvergence problem, increase the diversity of population, improve the convergence speed,and enhance the convergence efficiency, it is reasonable and effective.(2) Cloud fitness calibration is proposed in this paper. This method can make betterindividuals and other Individuals are similarly in fitness values, avoid few excellentindividuals filling with the whole group, and maintain the diversity of population in the initialstage; this method can enlarge the differences between the best individual and otherindividuals, avoid the individuals being bred to the next generation with the similarprobability, and increase the competition between individuals. In addition, this paper putsforward another cloud fitness calibration for disruptive selection. It gives higher fitness valuesto better individuals and poor individuals for the higher probability of reproducing to the nextgeneration, lower fitness values to the general individuals, and makes full use of theinformation in the poorer individuals. Experiment proves that the genetic algorithm using cloud fitness calibration can more easily find the optimal solution of the type of ’a needle inthe sea’ function.(3) Cloud initialization is proposed in this paper. In the condition of not knowing theexperiential knowledge of the solution space, it is difficult to determine the number of optimalsolutions and its distribution in the feasible solution space, so we often want to ensure evendistribution of initial population in the feasible solution space. However, the initial populationof normal genetic algorithm is randomly generated in general. This paper puts forward amethod of cloud initialization to uniformly sample in the feasible solution space, to makeinitial population have enough diversity. Simulation experiments show that the adoption ofcloud initialization in the algorithm avoids the premature convergence phenomenon to acertain extent, and enhances the global search ability of the algorithm.