Tuple leading differential evolution for black-box optimization

초록

Differential evolution (DE) has achieved giant success in finding globally optimal solutions to simple optimization problems. Nevertheless, it encounters numerous challenges when addressing increasingly complicated optimization problems. To promote its effectiveness in addressing such optimization problems, this paper designs a tuple leading differential evolution (TLDE) algorithm. Concretely, a tuple-based competition mechanism is first designed for TLDE to randomly partition individuals into three sets, namely two champion sets and one non-champion set, by using two different random tuple sizes. Then, two heterogeneous mutation schemes are devised to update individuals in the two champion sets and the non-champion set, respectively. With these two techniques, relatively better individuals orient the mutation of relatively worse ones. Resultantly, individuals in the champion sets pay more attention to intensively mining the optimal regions they lie in to seek high-quality solutions, while those belonging to the non-champion set devote themselves more to cruising the problem space in a variety of directions with the attempt to locate as many promising regions as possible. In this way, TLDE expectedly compromises search convergence and search diversity appropriately to seek high-precision solutions. At last, abundant experiments are performed on the CEC’2017 suite with four dimensionality configurations and the CEC’2022 suite with two dimensionality settings to compare TLDE with totally 16 state-of-the-art optimization methods including 11 latest DEs and 5 latest other metaheuristic algorithms. The experimental results reveal that TLDE attains considerably competitive performance or even prominently superior performance to the 16 compared algorithms. Further comparisons between TLDE and the CEC’2017 and the CEC’2022 winners uncover that the designed mutation scheme in TLDE helps all the winners achieve notably superior performance. Additional experiments on CEC’2011 practical unconstrained problems and three engineering design problems further reveal that TLDE is particularly effective in addressing real-world problems. In particular, all experimental results reveal that TLDE preserves an outstanding scalability in tackling optimization problems with higher dimensionalities and especially is adept at coping with complicated optimization problems. © 2025 Elsevier Ltd

키워드