.png){kind=icon}
.png){kind=icon}
Return
A large neighborhood search algorithm based on Q-learning for multi-container loading problem
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.126429.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The multi-container loading problem is a practical problem commonly faced by manufacturing and logistics industries. Solving this problem can effectively improve transportation efficiency and reduce logistics costs. In the current production process, the strong heterogeneity of items, containers and the large order size pose significant challenges to the generation of loading solutions. In this study, we propose a heuristic packing algorithm and a novel Q-learning-based large neighborhood search (QLNS) algorithm to solve this problem in acceptable time. Specifically, the items are selected and packed into containers based on the heuristic packing algorithm, and the tail-packing strategy is designed to optimize the criterion for container selection in the final stage of packing. The QLNS algorithm is proposed to improve the initial loading solution generated by the heuristic packing algorithm. Unlike traditional random or adaptive selection of operators, this algorithm designs three features as the state of the solution, and chooses a pair of destroy-repair operators under the guidance of a trained agent. Compared with the other two approaches, this approach is more targeted to improve the loading solution. In addition, a prioritized experience replay mechanism is introduced to assist in the training of the agent. Experiments on real-world instances demonstrate that our algorithms can obtain satisfactory loading solutions for different scales of items within a specified time and outperforms other existing large neighborhood search algorithms.
Keywords:
Multi-container loading
Q-learning
Large neighborhood search
Heuristic packing algorithm
Average loading rate