An overview of parameter control and adaptation strategies in differential evolution algorithm
Ke Tang · Caai Transactions on Intelligent Systems · 2011
Differential evolution algorithms have gradually become one of the most popular types of stochastic search algorithms in the area of evolutionary computation.They have been successfully applied to solve various problems in real-world applications.Since their performance often depends heavily on the parameter settings,the design of parameter control and adaptation strategies is one of the current hot topics of research in differential evolution.Although numerous parameter control schemes have been proposed,systematic overviews and analysis are still lacking.In this paper,first the basic principles and operations of the differential evolution algorithm were briefly introduced,and then a detailed overview was provided on different parameter control and adaptation strategies by dividing them into the following four classes: empirical parameter settings,randomized parameter adaptation strategies,randomized parameter adaptation strategies with statistical learning,and parameter self-adaptation strategies.The overview emphasized the latter two classes.To study the efficacy of these parameter control and adaptation strategies,experiments with the background of real-valued function optimization were conducted to compare their efficiency and practicability further.The results showed that the parameter self-adaptation is one of the most effective strategies so far.