A dynamic task-assisted constrained multimodal multi-objective optimization algorithm based on reinforcement learning

•A dynamic task-assisted method is designed to solve constrained multimodal problems.•Three auxiliary tasks evolve cooperatively.•Dynamic selection of auxiliary tasks with reinforcement learning.•A new indicator for integrated evaluation of the quality of solutions is designed. Constrained multimoda...

Full description

Saved in:
Bibliographic Details
Published in:Swarm and evolutionary computation Vol. 98; p. 102087
Main Authors: Wang, Zheng, Ye, Qianlin, Wang, Wanliang, Li, Guoqing, Dai, Rui
Format: Journal Article
Language:English
Published: Elsevier B.V 01.10.2025
Subjects:
Constrained multimodal optimization
Evolutionary algorithms
Muti-task
Reinforcement learning
Evolutionary algorithms
Muti-task
Constrained multimodal optimization
Reinforcement learning
ISSN:2210-6502
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A dynamic task-assisted method is designed to solve constrained multimodal problems.•Three auxiliary tasks evolve cooperatively.•Dynamic selection of auxiliary tasks with reinforcement learning.•A new indicator for integrated evaluation of the quality of solutions is designed. Constrained multimodal optimization problems (CMMOPs) are required to satisfy constraint limitations and ensure the convergence and diversity of the solutions in the objective and decision spaces. It increases the difficulty of solving the optimization problems. To design efficient constrained multimodal multi-objective optimization evolutionary algorithms (CMMOEAs) to solve them is a hot topic today. A novel dynamic auxiliary task selection algorithm (DTCMMO-RL) is designed based on the multi-task framework and reinforcement learning. The algorithm designs three auxiliary tasks to optimize constrained multi-objective problems, simple multi-objective problems and multimodal optimization problems, respectively. At the same time, Q-learning in reinforcement learning is employed to dynamically select the current optimal auxiliary task to utilize the useful information obtained rationally. In addition, an indicator (IGDXp) capable of evaluating the comprehensive performance of the solutions in the objective space and decision space is designed. To verify the excellence of DTCMMO-RL, a series of experiments with 11 comparison algorithms on CMMF and CMMOP are conducted to verify the feasibility and effectiveness of multiple strategies.
ISSN:2210-6502
DOI:10.1016/j.swevo.2025.102087