.png){kind=icon}
.png){kind=icon}
Return
An improved multi-operator differential evolution via a knowledge-guided information sharing strategy for global optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.126403.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Although the improved multi-operator differential evolution (IMODE) has won the CEC2020 competition, it still has the drawback of the randomness and uncertainty due to its use of a random information sharing strategy between sub-populations. Therefore, it is not easy for the mutation operator to evolve using the promising sub- population, and its strength is hard to be fully utilized especially for complex problems. Herein, we observe that the ability of the information sharing strategy to learn the evolutionary knowledge plays a crucial role in the issues raised. To this end, this paper proposes an enhanced IMODE, called IMODE-KG, via a novel knowledge- guided information sharing strategy. Our proposal consists of three main processes, including agent guidance, independent learning and knowledge update. In agent guidance, the reinforcement learning (RL) agents guide the algorithm to sample the sub-populations from the entire population according to the results of the fitness landscape analysis. Then, in independent learning, each mutation operator independently evolves its sampled sub-population for certain generations. In knowledge update, the agent updates its knowledge by learning the fitness landscape features of the problem in the previous process, to ensure the accuracy of the guidance. The superior performance of IMODE-KG is verified against state-of-the-art competitors on the CEC2020, CEC2021 test sets and fifteen engineering design problems.
Keywords:
Differential evolution
Knowledge-guided
Information sharing strategy
Engineering design problems
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W