The Improvement Of Differential Evolution Algorithm And Its Application

As a kind of relatively new global optimization evolution Algorithm,Differential Evolu-tion Algorithm[1](referred to as "DE”), is widely used because of its good convergence,simple model,easy to implement and less control parameters since1995when it is proposedby Rainer Storn and Kenneth Price. Therefore, Differential Evolution algorithm stands outfrom various evolution algorithms and it has developed very rapidly in recent years.So moreand more attentions are paid to it and its application is becoming more and more widely.This paper introduces the basic principle of Differential Evolution Algorithm, detailedimplementation process, parameter control study and its effect on the performance ofalgorithm. Compared with other evolution algorithms, Differential Evolution Algorithm hasmany advantages, but its shortcomings are also cannot be ignored, such as the conventionaldifferential evolution algorithm may be trapped in local optimum and stagnation phenomenonexists, which make the algorithm stop to converge. In view of this, this paper puts forward theimproved Differential Evolution Algorithm. Both the improvement of control parameters andthe improved differential evolution strategy are included. Test it with the benchmark functions,the result shows the effectiveness of the improvements.The modification not only preventsthe stagnation phenomenon, speeds up the convergence rate, but also makes the algorithmconverge to the global optimal solution.Traditional K-means algorithm is sensitive to the choice of initial clustering center, easyto fall into local optimum and the value of K is difficult to determine. Differential Evolutionalgorithm is a kind of heuristic global search techniques based on population with real numberencoding and its optimization on real value parameters has very strong robustness. This paperputs forward a K-means clustering algorithm based on Differential Evolution algorithm inorder to overcome the shortcomings. The method takes K as a parameter and combines theefficiency of K-means clustering algorithm with the global optimization ability of DifferentialEvolution algorithm. At the same time we insert K-means clustering algorithm to the newindividuals to speed up the convergence, so that it can solve the optimization of clusteringcenter better, find the best K and we can realize the goal to improve traditional clusteringmethods. The experiment results show that this algorithm can effectively improve theperformance of clustering algorithms.