Search for a command to run...
Purpose Although Differential Evolution (DE) algorithm has significant advantages in solving complex optimization problems in the real world, premature convergence and trapping into local optima seriously reduce its optimization performance. Thus, this study aims to improve the optimization performance of the DE by proposing a novel adaptive DE algorithm with local optimum region marking strategy (MDE). Design/methodology/approach MDE introduces three improvement strategies to enhance the optimization performance. A sine function that simulates chaotic sequence variations, incorporating historical information to self-adaptive adjust scale factor and crossover rate. The mutation operator DE/current-to-pbest/1 is improved by using an inverse roulette wheel selection enabling MDE to maintain a good balance between exploration and exploitation when addressing complex problems. Inspired by the behavior of people playing hide-and-seek games in the real world, a local optimum region marking strategy is proposed to improve the probability of MDE finding the global optimum while helping it maintain its exploratory ability throughout the evolution process. Findings The optimization performance of MDE is evaluated using test functions from the CEC2017 benchmark suite and compared with five advanced DE variants, demonstrating that MDE exhibits competitive optimization performance on continuous optimization problems. The MDE is additionally tested on the traveling salesman problem, with results indicating its significant potential for solving combinatorial optimization problems. Originality/value This study introduces the novel DE variant named MDE, providing an effective tool for solving complex optimization problems. The core contribution of this work lies in proposing the novel local optimum region marking strategy that effectively mitigates the impact of premature convergence and trapping into local optima on MDE. Moreover, the improvement strategies proposed can be easily transferred to other DE variants, and we believe they have the potential to enhance the optimization performance of those variants as well.
Published in: International Journal of Intelligent Computing and Cybernetics
Volume 19, Issue 1, pp. 55-85