Improvement Research Of Differential Evolution Algorithm

Differential Evolution algorithm(DE)is a new intelligent optimization algorithm proposed by American scholars Storn and Price in 1995.With advantages such as fewer parameters,simple operation principle and good convergence performance,DE has been widely used in evolutionary computing.Although differential evolutionary algorithm has its unique advantages,it can not completely get rid of the defects of intelligent evolutionary algorithm such as search stagnation and sensitivity to control parameters.In view of these problems,this thesis proposes two improved DE algorithms,and validates the effectiveness of the improved algorithm through standard test functions.The main contents of this thesis are as follows:Firstly,the development history of evolutionary computing is briefly introduced,and both the advantages and disadvantages of DE algorithm are listed.Also,as there are many improved algorithms in recent years,some excellent ones are sorted out,and the main improvement methods and application directions of DE algorithm are summarized.The operation flow of DE algorithm is introduced in detail,and the improvement of differential evolution algorithm is discussed in four aspects: population structure,control parameters,strategy selection and hybrid algorithm.Secondly,with the purpose of solving the hot problem of large-scale global optimization,a multi-populations differential evolution algorithm based on optimal random mutation strategy is proposed.The whole population is divided into several sub-populations according to different fitness values of each individual,which makes superior individuals go faster in its evolution.Not only did it improve the overall convergence rate of the algorithm,but also improved the late evolution of the population diversity.In order to keep the balance between search ability and development ability of the algorithm,it also introduces a new mutation strategy to generate basis vectors by linear combination of optimal individuals and random individuals.The performance of the improved algorithm is tested on a set of 19 large-scale continuous optimization problems.The results show that the improved algorithm is capable of solving large scale global optimization problems.Thirdly,in order to further improve the convergence performance of the algorithm,an adaptive multi-strategies differential evolution algorithm is proposed,which adopts a new adaptive mechanism,which dynamically adjusts the mean value of the normal distribution according to the number of iterations.Three different mutation strategies,"DE/rand/1","DE/rand assemble/1" and "DE/best/2",are applied to three sub-populations,in which the base vector of "DE/rand assemble/1" is composed of three random individuals proportionally.The difference vector is composed of random individuals who are different from the basisvector.Finally,the algorithm is evaluated on 25 benchmark functions of CEC 2005 and is compared with several other advanced improved algorithms.The results show that the improved algorithm has faster convergence speed and better optimization ability.