Research On Information Extraction Strategy Based Differential Evolution Algorithm

Evolutionary algorithms(EAs),inspired by the biological evolution of nature,are computationally robust methods that could adapt in different environment and different problems to give satisfactory solutions.Differential evolution(DE)algorithm is a very popular EA.In the traditional differential mutation strategy of DE,individuals in the population are randomly selected as a parent vector for mutation,which results in the drawback of blindness.To solve this problem,many scholars began to study the genetic evolution of the best individuals in the current population,as well as to improve the performance of the algorithm based on the optimal individual's neighborhood information,and achieved good research results.However,these evolutionary strategies are often time-consuming.In order to remedy the blindness of evolution in the traditional differential evolution algorithm,a framework based on information extraction strategy(IES)is proposed in this paper.The framework is applied to the mutation operator.The main work and innovation of this article are summarized as follows:By synthetically analyzing the main idea of evolutionary computation and the mathematical rules of information extraction,IES is proposed to improve the performance of the algorithm by using the extraction of valid information between individuals in a population to guide the evolutionary direction.Since IES is generic,it is not limited by the type of evolutionary algorithm and can be easily applied to different mutation operators.After verifying the effectiveness of two individuals based IES,this paper further investigates the optimal number of individuals that participate in IES.Numerical experiments show that the proposed strategy significantly enhances the performance of classic differential evolution algorithm as well as several advanced differential evolution variants.Moreover,IES outperforms the ranking based mutation scheme.Meanwhile,the proposed IES strategy also exhibits superior performance and applicability in several real-world optimization problems.