TY - JOUR
T1 - Learning-based logistics planning and scheduling for crowdsourced parcel delivery
AU - Kang, Yuncheol
AU - Lee, Seokgi
AU - Chung, Byung Do
N1 - Funding Information:
This work was supported by the Hongik University new faculty research support fund (2016S142401).
Publisher Copyright:
© 2019 Elsevier Ltd
PY - 2019/6
Y1 - 2019/6
N2 - Today many domains have begun dealing with more complex and practical problems thanks to advances in artificial intelligence. In this paper, we study the crowdsourced parcel delivery problem, a new type of transportation, with consideration of complex and practical cases, such as multiple delivery vehicles, just-in-time (JIT)pickup and delivery, minimum fuel consumption, and maximum profitability. For this we suggest a learning-based logistics planning and scheduling (LLPS)algorithm that controls admission of order requests and schedules the routes of multiple vehicles altogether. For the admission control, we utilize reinforcement learning (RL)with a function approximation using an artificial neural network (ANN). Also, we use a continuous-variable feedback control algorithm to schedule routes that minimize both JIT penalty and fuel consumption. Computational experiments show that the LLPS outperforms other similar approaches by 32% on average in terms of average reward earned from each delivery order. In addition, the LLPS is even more advantageous when the rate of order arrivals is high and the number of vehicles that transport parcels is low.
AB - Today many domains have begun dealing with more complex and practical problems thanks to advances in artificial intelligence. In this paper, we study the crowdsourced parcel delivery problem, a new type of transportation, with consideration of complex and practical cases, such as multiple delivery vehicles, just-in-time (JIT)pickup and delivery, minimum fuel consumption, and maximum profitability. For this we suggest a learning-based logistics planning and scheduling (LLPS)algorithm that controls admission of order requests and schedules the routes of multiple vehicles altogether. For the admission control, we utilize reinforcement learning (RL)with a function approximation using an artificial neural network (ANN). Also, we use a continuous-variable feedback control algorithm to schedule routes that minimize both JIT penalty and fuel consumption. Computational experiments show that the LLPS outperforms other similar approaches by 32% on average in terms of average reward earned from each delivery order. In addition, the LLPS is even more advantageous when the rate of order arrivals is high and the number of vehicles that transport parcels is low.
KW - Admission control
KW - Continuous feedback variable control
KW - Crowdsourced parcel delivery
KW - On-demand delivery service
KW - Reinforcement learning
UR - http://www.scopus.com/inward/record.url?scp=85064740895&partnerID=8YFLogxK
U2 - 10.1016/j.cie.2019.04.044
DO - 10.1016/j.cie.2019.04.044
M3 - Article
AN - SCOPUS:85064740895
SN - 0360-8352
VL - 132
SP - 271
EP - 279
JO - Computers and Industrial Engineering
JF - Computers and Industrial Engineering
ER -