A hybrid approach based on tabu search and reinforcement learning for container loading problems

Container loading problem holds a significant place in the logistics sector, playing a critical role in reducing shipping costs and increasing operational efficiency. This problem is a combinatorial optimization problem that involves placing three-dimensional rectangular items into a container in the most efficient way under certain constraints, and it belongs to the NP-hard class. In this study, a novel hybrid algorithm (TSQL) based on tabu search and reinforcement learning was developed for container loading problems. The aim of the study is to contribute to the literature by developing a new method and to provide a method that can reduce logistics costs in industrial applications and benefit the national economy and social welfare by producing quick and effective solutions to the real-world problem of container loading. Container loading problems were addressed in two stages. The first stage involves optimizing the loading sequence of the box types to be loaded into the container, and the second stage involves placing the boxes into the container in the most efficient way according to the determined sequence. The loading sequence of the box types into the container is determined by the hybrid algorithm (TSQL) based on tabu search and reinforcement learning. In the TSQL algorithm, the Q-learning algorithm is integrated into the 'generation and evaluation of neighbors' stage of the tabu search method. Additionally, a placement heuristic is used to place the items in the container most efficiently. Within the scope of the thesis, a hybrid model that uses these three methods together was created, and its performance was tested on comparative datasets. The findings show that the developed hybrid algorithm can produce faster and higher quality solutions compared to the tabu search method. This study represents a significant innovation as it is the first study in the literature to use Q-learning to enhance the tabu search algorithm, and it is also the first study in the container loading problems literature to use a hybrid approach combining a metaheuristic algorithm with reinforcement learning.