The first mile is the hardest: A deep learning-assisted matheuristic for container assignment in first-mile logistics

TY - JOUR

T1 - The first mile is the hardest: A deep learning-assisted matheuristic for container assignment in first-mile logistics

AU - Emde, Simon

AU - Tudoran, Ana Alina

PY - 2025/1/24

Y1 - 2025/1/24

N2 - Urban logistics has been recognized as one of the most complex and expensive part of e-commerce supply chains. An increasing share of this complexity comes from the first mile, where shipments are initially picked up to be fed into the transportation network. First-mile pickup volumes have become fragmented due to the enormous growth of e-commerce marketplaces, which allow even small-size vendors access to the global market. These local vendors usually cannot palletize their own shipments but instead rely on containers provided by a logistics provider. From the logistics provider's perspective, this situation poses the following novel problem: from a given pool of containers, how many containers of what size should each vendor receive when? It is neither desirable to supply too little container capacity because undersupply leads to shipments being loose-loaded, i.e., loaded individually without consolidation in a container; nor should the assigned containers be too large because oversupply wastes precious space. We demonstrate NP-hardness of the problem and develop a matheuristic, which uses a mathematical solver to assemble partial container assignments into complete solutions. The partial assignments are generated with the help of a deep neural network (DNN), trained on realistic data from a European e-commerce logistics provider. The deep learning-assisted matheuristic allows serving the same number of vendors with about 6% fewer routes than the rule of thumb used in practice due to better vehicle utilization. We also investigate the trade-off between loose-loaded shipments and space utilization and the effect on the routes of the collection vehicles.

AB - Urban logistics has been recognized as one of the most complex and expensive part of e-commerce supply chains. An increasing share of this complexity comes from the first mile, where shipments are initially picked up to be fed into the transportation network. First-mile pickup volumes have become fragmented due to the enormous growth of e-commerce marketplaces, which allow even small-size vendors access to the global market. These local vendors usually cannot palletize their own shipments but instead rely on containers provided by a logistics provider. From the logistics provider's perspective, this situation poses the following novel problem: from a given pool of containers, how many containers of what size should each vendor receive when? It is neither desirable to supply too little container capacity because undersupply leads to shipments being loose-loaded, i.e., loaded individually without consolidation in a container; nor should the assigned containers be too large because oversupply wastes precious space. We demonstrate NP-hardness of the problem and develop a matheuristic, which uses a mathematical solver to assemble partial container assignments into complete solutions. The partial assignments are generated with the help of a deep neural network (DNN), trained on realistic data from a European e-commerce logistics provider. The deep learning-assisted matheuristic allows serving the same number of vendors with about 6% fewer routes than the rule of thumb used in practice due to better vehicle utilization. We also investigate the trade-off between loose-loaded shipments and space utilization and the effect on the routes of the collection vehicles.

KW - logistics

KW - machine learning

KW - Assignment

KW - First-mile logistics

KW - Deep neural network

KW - E-commerce supply chain

KW - Matheuristic

UR - http://www.scopus.com/inward/record.url?scp=105001701079&partnerID=8YFLogxK

U2 - 10.1016/j.ejor.2025.01.024

DO - 10.1016/j.ejor.2025.01.024

M3 - Journal article

SN - 0377-2217

VL - 324

SP - 335

EP - 350

JO - European Journal of Operational Research

JF - European Journal of Operational Research

IS - 1

ER -