Self-adaptive differential evolution algorithms based on complementary mutation operators

Based on the differential evolution (DE) algorithm with self-adaptive parameters, several novel DE algorithms are proposed. These algorithms adopt both DE/rand/1 and DE/best/2 mutation operators which are of complementary virtue, and adjust the use of two mutation operators by multiple different assignment strategies including random assignment, monotone assignment based on population size, adaptive random assignment and adaptive assignment based on population size. The results of numerical optimization based on benchmark test functions show that the self-adaptive DE algorithms with two mutation operators are obviously better than two canonical DE algorithms. Among the four assignment strategies, the monotone assignment has the best effect. The proposed DE algorithms take advantage of the DE/rand/1 mutation in preserving population diversity, inherit the virtue of fast local convergence from the DE/best/2 mutation and achieve a better tradeoff between exploration and exploitation. Moreover, in these new DE algorithms few parameters need to be tuned manually, so it is convenient to use them in practice.