Attention-Reinforcement Learning-Based Adaptive Genetic Algorithms for Order-Picking System

The Order-Picking System is vital to warehouse operations, directly affecting order fulfillment efficiency and cost management. Key processes such as order assignment, batching, and sequencing are optimized to minimize delays. With Industry 4.0 demanding real-time adaptability, traditional methods like Genetic Algorithms struggle with complex order patterns, leading to the need for more advanced techniques. This study introduces a Hybrid Reinforcement Learning-based Adaptive Genetic Algorithm (RLAGA) and its enhanced version, Attention-RLAGA, for optimizing order batching. RLAGA combines reinforcement learning with genetic algorithms to adaptively refine batch configurations, dynamically adjusting genetic operators. Attention-RLAGA builds on this by incorporating an attention mechanism to prioritize critical orders. Experimental results show that RLAGA improves computational efficiency by 76% and reduces total tardiness by over 5%, converging 93% faster than traditional GA-based methods. While Attention-RLAGA achieves marginally better tardiness reduction, it increases computational costs by over 400%, making it less practical for real-time use. RLAGA consistently performs well across different order sizes, and though RAILS achieves slightly lower tardiness, it requires 150% more time, highlighting a trade-off between efficiency and adaptability. RLAGA emerges as the most computationally efficient method, ideal for large-scale logistics, while Attention-RLAGA offers slight improvements at the cost of excessive computational demands. Both outperform GA-based heuristics in scalability, adaptability, and solution quality. This research advances reinforcement learning in genetic algorithms, providing a scalable framework for optimizing logistics in Industry 4.0. Future work will enhance RLAGA’s real-time adaptability and explore dynamic hybrid models for improved scalability and efficiency.