.png){kind=icon}
Return
Multi-objective differential evolution algorithm based on partial reinforcement learning intelligence for engineering design problems and physics-informed neural networks
.png){kind=icon}
PRE
AI
DOI：10.1016/j.aei.2026.104608.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes an innovative Multi-Objective Partial Reinforcement Learning Evolutionary Algorithm (MODEPRL). Unlike existing reinforcement learning-based DE variants that primarily focus on strategy selection or parameter adaptation, MODEPRL simulates intelligent behaviors observed in human or animal social groups through partial reinforcement learning (PRL). The core innovation lies in associating learners’ decision-making behaviors with adaptive score tables, enabling individuals to autonomously adjust their actions based on environmental stimuli defined by positive and negative reinforcement. This mechanism allows learners to collaboratively enhance the overall optimization performance. Numerical simulations on 27 mechanical design problems and physics-informed neural networks demonstrate that MODEPRL outperforms state-of-the-art algorithms in global search capability, solution accuracy, and convergence speed.
Keywords:
Multi-Objective Differential Evolution
Partial Reinforcement Learning
Adaptive Score Table
Engineering Design Problems
Physics-Informed Neural Networks
Journal
IF:
9.9
Papers:
4.4K
Citations:
1.7W
Organization
Cited Papers
.png){kind=icon}
ShareSave
ShareSaveShareSaveShareSave
View more