Improving differential evolution by differential vector archive and hybrid repair method for global optimization

Differential evolution (DE) has been extensively studied in the past decade, though researchers may ignore the effect of an archive storing previous search information. Moreover, boundary repair issue is seldom handled in the literature. This paper attempts to improve the performance of DE algorithm from these two aspects. First, a differential vector archive is constructed and adaptively updated during the optimization process of DE. The archive stores a set of differential vectors representing potential good search directions. Second, inspired by recently reported results about repair methods, a hybrid of four commonly used repair methods is proposed. The hybrid method is more applicable to unknown optimization problems than a single repair method. A test suite containing 28 benchmark functions is employed for experimental investigation. Experimental results show that the proposed algorithm usually affects the search to attain better performance in the later evolutionary stage. Our algorithm significantly outperforms a state- of-the-art algorithm. This result verifies the effectiveness of the proposed algorithm.