Retrieval-oriented storage relocation optimization of an automated storage and retrieval system

Business-to-consumer platforms (e.g., Tmall.com and JD.com) demand the efficient retrieval handling of storage systems. The amount of a certain item to be retrieved is often less than a pallet load; thus, dynamically re-assigning the storage locations of nonempty pallets after each retrieval may improve retrieval operational efficiency. We call such a dynamic decision on the location for the returned pallets relocation. We treat the relocation of each non-empty pallet as an operational decision rather than a tactical decision. Due to its dynamic nature, we formulate relocation as a dynamic program that minimizes the total crane travel time. To overcome the computational curse of dimensionality, we propose several heuristic relocation policies that are inspired by conventional storage location assignment policies. Based on the findings from preliminary numerical experiments, we design an approximate dynamic programming-based relocation policy that achieves a greater operational efficiency improvement than the aforementioned heuristics. Through a more extensive set of numerical experiments, we find that random relocation is generally no better than no relocation, and that closest-open relocation is effective across different rack shapes, crane velocity configurations, and retrieval characteristics. Our experiments also show that the approximate dynamic programming-based relocation policy consistently yields further improvement over closest-open relocation, at the expense of higher (but still affordable) computational complexity.