Dispatch Policies between Hubs in the Physical Internet with Emission Considerations and Local Dispatch Using Machine Learning

Abstract

This thesis presents a comprehensive analysis of dispatch management in Physical Internet (PI) logistics systems. As a modular approach to freight deliveries and container transportation, it aims to reduce costs and environmental impacts. The study focuses on dispatch management between PI hubs, particularly examining consolidation policies, traffic flow control, and managing stochastic arrivals. We employ two mixed-integer linear programming (MILP) models for local pairwise and centralized dispatch agents to conduct the experiments. A significant aspect of this research is the application of machine learning (ML) techniques, including Random Forest (RF), a multi-output regressor using RF (MReg-RF), and Artificial Neural Networks (ANNs), to enhance dispatch management within the PI framework. Our findings demonstrate substantial differences in efficiency and sustainability across various dispatch strategies and dispatch agent configurations. These results also highlight the pivotal role of ML in capturing the behavior of MILP models and managing dispatches with uncertainties in arrivals within the PI framework. The results offer valuable insights into integrating ML in PI logistics, presenting practical strategies to improve operational efficiency and sustainability.