| dc.description.abstract | In recent years, after COVID-19 pandemic, people tend to change their shopping habit
from offline in brick-and-mortar stores to online on websites and platforms due to its
convenience and their hectic learning and working schedule. As a result, a series of
businesses have also gradually switched to online business, causing the burden to
logistics activities, one of which is last-mile delivery. This study proposed a scenario
where a company not only own dedicated drivers (DD) with a pre-determined capacity
and availability for deliveries, has the option to utilize occasional drivers (OD), who are
willing to use their own vehicles for deliveries to exchange a small compensation fee.
This model is called “Crowd-shipping”, which can help the business to optimize the
total traveling cost for last-mile delivery. Besides the OD option, the business having
different status of customer orders was also taken into the consider in this research to
capture the reality. Since this focus was in the context of a stochastic and dynamic last mile delivery setting where the status of customer orders and the availability of OD
randomly appear through a day within a specific time windows for deliveries. To solve
this problem, a Deep Reinforcement Learning with Two-stage approach (DRL) under
Proximal Policy Optimization (PPO) was introduced to enable effective handling of
large-scale and real-life cases in last-mile delivery. Then, a capacitated vehicle routing
was build to develop further in defining the optimal routes for OD. The aim of this thesis
was to minimize the total cost for last-mile delivery with assigned allocation and routes
for two type of drivers. To demonstrate the effectiveness of the proposed system, a case
study involving an online platform was used to implement in the IBM CPLEX and
PYTHON software and its solution was conducted sensitivity analysis to evaluate the
performance. | en_US |
