Abstract: The selection of the parameter itself is a combinatorial optimization problem.Although a considerable number of works have been conducted,it is known to be a puzzled task.In this paper,a DE algorithm was proposed that uses a new mechanism to parameter self-selection,which dynamically learns from their previous experiences and selects the best performing combinations of parameters for the next generation during the convergence process.We firstly designed the mechanism including three aspects:building of parameter database,score of parameter performance and selection of parameter combination,then we conducted the experiments on some benchmark functions to judge the performance.The results show that the DE with the new mechanism obtains promising performance.

Key words: Differential evolution,Parameter self-adaptation,Parameter selection