An adaptive mutation strategy correction framework for differential evolution

Differential evolution (DE) is an efficient global optimization algorithm. However, due to its random properties, some individuals may mutate in the direction of deviating from the theoretical global optima, failing to evolve and wasting a lot of computing resources. Moreover, there is an imbalance between exploration and exploitation in mutation strategies. For these shortcomings, we propose an adaptive mutation strategy correction framework (AMSC) for DEs. In this framework, the population is firstly split into superior subpopulation and disadvantaged subpopulation. Two types of auxiliary mutant vectors based on the direction information are designed to respectively enhance the exploration ability and exploitation ability of these two subpopulations, so as to improve the search efficiency. Moreover, for achieving the proper balance between exploration and exploitation in DEs, we propose an adaptive cooperative rule for the above two auxiliary vectors based on the actual crossover rates. This rule controls the relative size of two subgroups to determine the proportion of two types of auxiliary vectors used in the whole population. To evaluate the performance of AMSC framework, we have introduced into eight original DEs and carried out comparative experiments on four practical problems and 59 test functions from CEC 2014 and CEC 2017 benchmark suites. The experiments demonstrate that the AMSC framework can increase DEs’ performance dramatically.