Answers are grounded in this paper and cite their sources.
Metaheuristic optimization algorithms are continuously facing challenges in exploration and exploitation throughout the search. The D optimization algorithm is inspired by neurobiological processes involved in memory consolidation during sleep, which includes three operators of exploitation, exploration, and integration. Nevertheless, the primal implementation applies these operators to all individuals sequentially, regardless of the location in the search space and optimization phase. This paper presents a novel hybrid algorithm called DOA-RL, which integrates Q-learning-based reinforcement learning into DOA to achieve adaptive operator selection. The approach defines the operator selection process as a Markov Decision Process. In this case, the state is defined as a combination of four important measurements: normalized fitness, fitness improvement, distance to best solution, and population diversity. The action space is defined as three DOA operators. The extensive experiments conducted on all 29 functions of the CEC 2017 benchmark indicate that the DOA-RL outperforms all with a win rate of 96.55% and won on 28 out of 29 functions. The algorithm yields an overall average improvement of 82.76%, with 72.44% on unimodal functions, 85.75% on hybrid functions, and 92.39% on composition functions. All winning functions are significantly better than the comparative algorithms at the 0.05 level with 30 independent runs per function.
No takes yet. Share an insight, caveat, or question.
Manhrawy et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: