An adaptive hybrid differential evolution algorithm for single objective optimization

A number of efficient differential evolution (DE) algorithms have been proposed in recent years to deal with constrained single objective optimization problems. Past studies have indicated that the performance of such algorithms is largely affected by the choice of parameters e.g., mutation factor, crossover rate, mutation strategy and the type of crossover. A combination of these parameters may work out to be the best for a problem while resulting in poor performance for others. This paper introduces an adaptive hybrid DE algorithm (AH-DEa). The algorithm employs a binomial crossover in early stages of evolution for exploration, while an exponential crossover is employed for exploitation in later stages. In addition, the crossover rate (CR) is adaptively controlled based on the success of offspring/trial solutions generated. A local search is initiated from the best found solution to explore possibilities of further improvement. Results of the proposed algorithm are compared with existing state of the art algorithms on a set of 40 widely studied mathematical benchmarks and two shape matching problems. The benefits of adaptive CR selection are highlighted. The results indicate that the proposed algorithm is able to identify better or comparable results across the wide range of single objective optimization problems.