Improving the search accuracy of differential evolution by using the number of consecutive unsuccessful updates

Abstract

During the process of evolution, feedback information is frequently utilized to guide the following search and it has been proved to be very effective. However, the information of unsuccessful updates has not been fully utilized yet. Thus, it is important to exploit this information systematically, which will provide a reference for further researches. In this paper, we use the number of consecutive unsuccessful updates ciG to improve the search accuracy of differential evolution (DE) and propose a novel type of DE variant, called as consecutive unsuccessful updates-based DE (CUSDE). There are two operations in CUSDE. Firstly, a simple but efficient mutation strategy is developed, in which the base vector and terminal vector are selected based on the probability calculated from ciG. Secondly, a deletion mechanism is proposed to adaptively remove some inferior individuals with larger ciG values. Compared with other state-of-the-art DE variants on CEC 2005 and CEC 2017 benchmark sets, the experimental results confirm the superiority of CUSDE in improving the search accuracy.