An improved arithmetic optimization algorithm based on reinforcement learning for global optimization and engineering design problems

doi:10.1016/j.swevo.2025.101985

An improved arithmetic optimization algorithm based on reinforcement learning for global optimization and engineering design problems

2025-07-01

0

PRE

AI

DOI：10.1016/j.swevo.2025.101985

Original

Original request for help

Save

Abstract

En 中文

To overcome the shortcomings of the Arithmetic Optimization Algorithm (AOA) in solution accuracy and convergence speed, this paper proposes an improved approach based on reinforcement Q-learning and Random Elite Pool strategy (QL-REP-AOA). The algorithm constructs a state space based on the iteration process and designs a nonlinear reward function with stage adaptability. With this design, the algorithm can dynamically select the optimal search strategy based on the characteristics of each stage of the optimization problem. Additionally, the Random Elite Pool strategy is introduced, which enhances population diversity and search efficiency through the collaborative effect of multiple search operators. To validate the effectiveness of the proposed algorithm, experiments are conducted on 27 classical benchmark functions, the CEC2020 test set, and real-world engineering problems. The experimental results show that QL-REP-AOA outperforms other optimization algorithms in both accuracy and convergence speed, demonstrating its potential in solving complex optimization problems.

Keywords:

Q -learning algorithm

Random elite pool

Arithmetic optimization algorithm

Reward function

An improved arithmetic optimization algorithm based on reinforcement learning for global optimization and engineering design problems

Abstract

Journal

Organization