.png){kind=icon}
.png){kind=icon}
Return
Deep reinforcement learning assisted automated guiding vector selection for large-scale sparse multi-objective optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2024.101606.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Sparse multi -objective optimization problems (SMOPs) are prevalent in a wide range of applications, spanning from the fields of science to engineering. Existing sparse evolutionary algorithms utilize single or multiple guiding vectors to direct the generation of offspring solutions in a constant manner according to human experience, which are difficult to determine the best guiding vector for various population states and prone to falling into premature convergence. To address the dilemma in guiding vector adaptation, this paper proposes a novel guiding vector selection method based on reinforcement learning. In the proposed method, the features extracted from the current population are regarded as states, the overall degrees of improvement in population convergence and diversity are regarded as rewards, and the candidate guiding vectors are regarded as actions. By using deep neural networks to establish the mapping model between population states and the expected cumulative rewards of guiding vectors, the proposed method can determine the best guiding vector for the current population at each generation. The selected guiding vector inspires the development of novel genetic operators, which can approximate sparse Pareto optimal solutions in high -dimensional decision spaces. Experimental results on both benchmark and real -world SMOPs demonstrate that the proposed algorithm has when with the state-of-the-art.
Keywords:
Multi-objective optimization
Sparse optimization
Evolutionary computation
Deep reinforcement learning
Variable clustering
Journal
IF:
8.5
Papers:
2.1K
Citations:
1.0W