Interpretable and Risk‑Aware Decision‑Making in Rail-Truck Intermodal Dangerous Goods Transportation and Flow Estimation

Abstract

This dissertation investigates rail–truck intermodal transportation of dangerous goods, with particular emphasis on modeling and estimating their flows within highway networks. The dissertation is structured into five chapters.

Chapter 1 introduces the research context, motivation, and main contributions of the study. Chapter 2 presents the first study, which develops an alternative tactical-level approach for configuring existing rail–truck intermodal networks for dangerous goods shipments from the perspective of intermodal carriers. Transportation of dangerous goods involves inherently conflicting considerations of cost efficiency and public risk, representing the differing priorities of carriers and regulators. Rather than treating these priorities as competing objectives, the study formulates the problem as a bi-level optimization model that embeds stakeholder preferences within a unified decision-making framework. Extending the classical hub-and-spoke paradigm, the upper-level model represents carrier cost considerations through a single-allocation hub selection problem, while the lower-level model captures regulatory risk concerns. The proposed formulation contributes by (i) explicitly reconciling stakeholder objectives within a single model and (ii) simultaneously determining terminal selection and shipment routing. The model is evaluated using a real-world Canadian rail–truck intermodal network and benchmarked against a conventional bi-objective formulation. Case study results demonstrate improved solution quality and enhanced stakeholder alignment.

Chapter 3 extends this work by addressing two major challenges associated with large-scale combinatorial optimization models: computational tractability under parameter variability and interpretability of solutions as operating conditions change. The second study proposes a learning-to-optimize framework that extracts structural insights from historical optimization outcomes, thereby reducing reliance on repeated computationally intensive solver-based solutions. Developed for the configuration of rail–truck intermodal networks carrying dangerous goods, the methodology is demonstrated through a case study on a realistic U.S. intermodal network. Results show that the approach facilitates decision-making under uncertainty by translating complex optimization outputs into intuitive and interpretable decision rules.

Chapter 4 focuses on risk assessment in dangerous goods transportation by introducing a comprehensive data-driven, optimization-based framework to estimate dangerous goods flows on road networks. By integrating multiple data sources, the framework estimates and spatially distributes hazardous material flows using an optimization model, providing detailed exposure and vulnerability information for risk analysis and mitigation planning. A case study on gasoline transportation in Ontario, Canada, highlights spatial flow patterns and yields insights into accident dynamics involving hazardous materials, supporting more informed and proactive risk management strategies. Finally, Chapter 5 summarizes the main findings of the dissertation and outlines directions for future research.