Covariance matrix adaptation, evolutionary game theory and deep Q-Networks integrated evolutionary algorithm

Purpose Optimization is essential in business, engineering and AI-driven decision-making. However, conventional methods like Differential Evolution (DE) struggle with high-dimensional, dynamic and multi-modal problems due to fixed adaptation mechanisms. This study aims to introduce a hybrid metaheuristic approach, EvoQ-EA (Evolutionary Q-Learning-Based Evolutionary Algorithm), which integrates Covariance Matrix Adaptation – Evolution Strategy (CMA-ES), Evolutionary Game Theory (EGT) and Deep Q-Networks (DQN) to enhance adaptive search and decision-making. Design/methodology/approach EvoQ-EA dynamically selects mutation and crossover strategies using reinforcement learning (DQN), optimizes exploration-exploitation balance via CMA-ES and refines search space reduction through EGT-based historical solution analysis. The framework is evaluated on 13 benchmark test functions for robustness and tested on a real-world supply chain data set, benchmarking its performance against NSGA-II (Non-dominated Sorting Genetic Algorithm II). Additionally, EvoQ-EA is applied to Neural Architecture Search (NAS) and compared with NSGA-II and Random Search in terms of accuracy, computational efficiency and model size. Findings EvoQ-EA demonstrates superior robustness in supply chain optimization. It reduced the expected cost by 30.53% and the worst-case cost by 24.56% compared to NSGA-II. It improved the expected service level by 0.055% and the worst-case service level by 0.20%, highlighting its resilience under uncertainty. In NAS, EvoQ-EA outperforms NSGA-II and Random Search, achieving the highest accuracy (91% vs 88% and 85%), lowest computational cost (3.34 GFLOPs vs 8.11 and 7.89), but with a slightly larger model size (4.78 MB vs 2.47 and 2.92 MB). Originality/value This study bridges evolutionary algorithms with AI-driven decision-making, offering an adaptive, scalable and efficient solution for real-world optimization. By demonstrating superior robustness in supply chain management and efficiency in Neural Architecture Search, EvoQ-EA provides a next-generation approach to complex optimization challenges.