Research On Guided-clustering Multi-level Differential Evolution

Differential evolution(DE)algorithm is a new evolutionary algorithm for solving global optimization,and has been successfully applied to solve a series of engineering problems.But the differential evolution algorithm has some disadvantage,especially under the condition of limited number of the fitness function evaluation times to locate the global optimal solution in the global optimization.In real life,the clustering algorithm has a high frequency of application in the classification of information.In order to further improve the performance of evolutionary algorithm,we try to use the clustering algorithm to optimize the differential evolution algorithm.The main innovations of this paper are as follows:(1)Clustering-based differential evolution with composite trial vector generation strategies and control parameters(C-CODE)is proposed to get rid of the traditional poor weakness generation strategy parameters and trail vector single control differential evolution algorithm.And the clustering is more effective in using the population information,so as to improve the optimization performance and the robustness of the algorithm.The characteristics of difference trail vector generation strategies and the control parameters of DE have been extensively studied,which can be used to design more effective and more robust variants of DE.But in some adaptive DE and some of variants of the algorithm,for each target vector only the use of a trail vector generation strategy and a set of control parameters.C-CODE is proposed to break through limitations of search capability.(2)A guided-clustering differential evolution algorithm(GCDE)is proposed to overcome the random and blindness of traditional differential evolution algorithm based on clustering,not only based on the input data,and based on the fitness values in the process of guide clustering method.More special is that when the clustering process does not take into account the fitness value,and will be converted into the process of common clustering algorithm.And study through mathematical examples demonstrates the advantages of the proposed clustering algorithm.The simulation results on the international standard functions show that the differential evolution algorithm is superior to the differential evolution based on clustering in the optimization performance and convergence speed.(3)A guided-clustering multi-level differential evolution algorithm is proposed(GMDE).K-means algorithm is a typical clustering algorithm based on distance,but the result is better when the set is intensive and class difference is obvious.Based on this consideration.GMDE is proposed.When the population density using clustering method updates the population,whereas the traditional differential evolution algorithm is used.The superiority of the algorithm is proved by the simulation experiment on the international standard function.