Dynamic stochastic parcel locker assignment with uncertain pick-up times

Abstract

Automated parcel lockers are used by logistics providers in order to increase the efficiency of last-mile delivery operations particularly in urban areas. We consider the case of a company that operates lockers and dynamically receives delivery orders that have to be accepted or rejected immediately. On a second decision stage, the set of accepted orders has be assigned to the lockers such that customer compatibility requirements and maximum fulfillment times are respected. As both future arrivals of orders as well as customer pickup-times are unknown, the company faces a dynamic stochastic problem. To generate solutions, we propose a decision framework based on a classification approach. The classifier uses a mixed integer model to learn from optimal solutions within a deterministic setting and exploits this information within the dynamic stochastic process. We assess the proposed method within an extensive computational study where both artificial instances and a real world case are addressed. The obtained results show that the classification-based framework outperforms all benchmark methods, which include (i) scenario sampling, (ii) classical decision trees, and (iii) several deterministic policies. Managerial insights with regard to most important systems’ features within the decision process are derived. The newly proposed decision framework is generalizable such that it can be applied to related dynamic stochastic matching problems.