The Research On Improvement Of Differential Evolution Algorithm And Its Application

With the rapid development of modern industry and high-tech technology,optimization algorithms are becoming a hot spot in the study of scholars at home and abroad.In the specific industrial optimization field,the corresponding mathematical model of all process is established and the intelligent technique is applied in practical problems for productivity enhancement.As an self-heuristic and self-learning stochastic optimization method,the evolutionary algorithm is good at solving high-dimensional and multi-modal objective function,whose core idea is the evolution of nature species.Based on the principle of “the survival of fittest”,it is gradually improved its accuracy of global optimal solution through physical competition and crossover evolution between individual.The evolutionary algorithm simulates the competitive and survival behavior of population evolution process to find global optimal point of the objective function.At the some time,simulation tests by some researcher show that this algorithm preserves the fast stochastic convergence performance in various multiple hump functions.This paper mainly discuss the theory of differential vector model,also introduce some skill which is used to enhance the global searching ability,with attainment of high precision and fast convergence rate.After certain adjustments,the intelligent searching optimization technology is able to successfully solve some real production problems.Firstly,the paper introduces several branches of evolutionary algorithm in detail,makes a summary of its developing history and the newest research results,analyses the present academic research status and the background of differential evolution algorithm,point out future direction of its application and the trend of development.Secondly,the internal relations between selection operation of basic differential evolution and distribution of next generation information is systematically investigated,and the impact each parameter has on population search state is also analyzed.According to compare several experiment data,the optimal ranges of scale factor and crossover probability are suggested.The convergence of algorithms with different mutation strategies are compared.Furthermore,the advantages and disadvantages of the classical differential evolution algorithm are summarized.Aimed at differential evolution algorithm's defect of single searching mode,three improved methods are proposed based on the moving feature of individuals in different stages,roughly including the following several respect: Adjusting the size of scaling factor and the chosen probability of each dimensional information;Combining both DE/ rand/1 and DE/ best/1;The adaptive regulation of search step size.These novel strategies improve the individual's adaptability to the environment,and make the individual behaviors are easily controllable globally.So the above method keep tradeoff between local exploitation and the global exploration.In order to evaluate the optimization result using adjustment strategy,each algorithm which is applied to multiple benchamark functions,is showed that convergence accuracy of the adaptive searching algorithm is better than the existing evolutionary algorithms.In the basic differential evolution algorithm,the population diversity is gradually decreased at the latter evolution stage.By integrating different evolutionary algorithm,the hybrid optimization method prevent the phenomenon of falling into local optima.Under the existing method of hybrid algorithm and its inadequacy,three kind of intelligent optimal technology are combined: genetic,particle swarm and differential evolution algorithm,each of which has its optimization characteristic and updating way,and each species is applicable for a specific environment.Two subpopulation respectively adopt the difference evolutional optimization model and chromosomes genetic algorithm model.The individual,which has strong “random” and “purposive”search capability,after hybrid selection,use again standard particle swarm velocity update and location update operation.It is beneficial to direct offspring's search procedure according parent search experience in the feasible field,and the hybrid optimization method accomplishes dominance complement of the three evolutionary operations.The new hybrid algorithm has been tested on standard objective function and the application example,and numerical results show that the proposed algorithm has highest optimization accuracy and very practical value in application.