TY - JOUR
T1 - A learning-based approach for dynamic freight brokerages with transfer and territory-based assignment
AU - Min, Daiki
AU - Kang, Yuncheol
N1 - Funding Information:
This work was supported by Jungseok Logistics Foundation Grant.
Publisher Copyright:
© 2020 Elsevier Ltd
PY - 2021/3
Y1 - 2021/3
N2 - The recent evolution of information technology in logistics has facilitated the digital freight brokerage, which allows shippers and trucks to match needs and services in a short-term period. In the context of freight brokerages, this paper considers a dynamic fleet assignment problem related to matching demand and supply. We particularly integrate practical operational characteristics, such as territory-based assignment and transferring, which have not been considered in the dynamic fleet assignment problem. We first formulate the problem as a Markov Decision Process (MDP) to represent uncertain and sequential decision-making procedures. Furthermore, to overcome the dimensionality and ambiguity of the MDP model, we proposed a reinforcement learning(RL) approach with function approximation for solving the MDP model. Finally, numerical experiments are carried out to illustrate the superiority of the RL method and analyze the effects of operational characteristics. A numerical analysis shows that the proposed RL-based method provides more rewards than other policies, such as myopic policy and first-come-first-served (FCFS) policy, for all test scenarios. The RL-based method is even better for a situation in which the delivery territories are highly overlapped and customer demand exceeds to supply capacity, which requires precise capacity control. In addition, we observe that significant achievement can be attained by allowing trucks to deliver with transfer.
AB - The recent evolution of information technology in logistics has facilitated the digital freight brokerage, which allows shippers and trucks to match needs and services in a short-term period. In the context of freight brokerages, this paper considers a dynamic fleet assignment problem related to matching demand and supply. We particularly integrate practical operational characteristics, such as territory-based assignment and transferring, which have not been considered in the dynamic fleet assignment problem. We first formulate the problem as a Markov Decision Process (MDP) to represent uncertain and sequential decision-making procedures. Furthermore, to overcome the dimensionality and ambiguity of the MDP model, we proposed a reinforcement learning(RL) approach with function approximation for solving the MDP model. Finally, numerical experiments are carried out to illustrate the superiority of the RL method and analyze the effects of operational characteristics. A numerical analysis shows that the proposed RL-based method provides more rewards than other policies, such as myopic policy and first-come-first-served (FCFS) policy, for all test scenarios. The RL-based method is even better for a situation in which the delivery territories are highly overlapped and customer demand exceeds to supply capacity, which requires precise capacity control. In addition, we observe that significant achievement can be attained by allowing trucks to deliver with transfer.
KW - Dynamic assignment problem
KW - Freight brokerage
KW - Reinforcement learning
KW - Territory-based assignment
KW - Transferring
UR - http://www.scopus.com/inward/record.url?scp=85098721733&partnerID=8YFLogxK
U2 - 10.1016/j.cie.2020.107042
DO - 10.1016/j.cie.2020.107042
M3 - Article
AN - SCOPUS:85098721733
VL - 153
JO - Computers and Industrial Engineering
JF - Computers and Industrial Engineering
SN - 0360-8352
M1 - 107042
ER -