An integrated approach for lot-sizing and storage assignment

Abstract

In this paper, we study the interaction between the lot-sizing problem and the storage assignment problem. Traditional lot-sizing problems have been studied for decades. However, only recent studies have further considered decisions related to the assignment of items to inventory locations, aiming to better model the complex reality. In our problem, the storage space is divided into several separate locations, and the inventory is assigned to the storage locations taking into account specific compatibility conditions. Relocation of inventory is also possible if needed. In addition to the traditional cost elements from the lot-sizing problem, we consider others related to holding inventory, such as fixed storage costs, handling costs, and relocation costs. We model the problem using a general mathematical model, as well as a transportation reformulation, which provides better lower bounds. We propose several heuristics to solve the problem by splitting it into smaller subproblems, which are then solved sequentially. A series of computational experiments is carried out in order to evaluate the impact of the integration between the lot-sizing and the storage assignment decisions, as well as the behavior of the different solution approaches. The results show that the proposed heuristics are highly effective in finding feasible solutions that are very close to the best solutions, while spending 97% less computational time compared to solving the full mathematical model. When compared to the relax-and-fix heuristic (benchmark), certain versions of the heuristics can find better solutions using less computational effort, underscoring the benefit of employing more specialized heuristics. Additionally, we conduct a sensitivity analysis with the aim of understanding the impact of key input parameters on the problem. The results indicate a significant influence of compatibility levels on the problem complexity. Limited item–item compatibility notably increases complexity, whereas restricted item–location compatibility reduces computational time.