Date of Graduation

5-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Civil Engineering

Advisor/Mentor

Hernandez, Sarah V.

Committee Member

Eksioglu, Sandra D.

Second Committee Member

Mitra, Suman K.

Third Committee Member

Nachtmann, Heather L.

Keywords

Automatic Identification System; Commercial Waterways; Freight; Machine Learning; Sotchastic Optimization; Waterborne Commerce Statistics

Abstract

Freight transportation is a crucial component of the US economy, with truck, rail, and water contributing significantly to the Gross Domestic Product (GDP). However, challenges arise in maintaining infrastructure capacity to accommodate the growing tonnage and value of freight transportation. Waterways offer high efficiency and environmental friendliness, with the capacity to transport substantial cargo volumes. Utilizing waterways can alleviate bottlenecks in landside transportation, accommodating the rising tonnage of commodity flow in the US. Despite the advantages of waterway transportation, the data limitations hinder informed decision-making regarding port operation and infrastructure investment. The lack of granularity and timeliness in available data sources makes it challenging to understand local conditions and respond to changing trends. The goal of this study is to address these data limitations through the development of machine learning and stochastic optimization models. This dissertation addresses data limitations in commercial waterways through the development of stochastic optimization and machine learning models. Stochastic optimization models account for uncertainties in commodity demand and port operations, offering insights for optimal investment strategies and supply chain management. Machine learning models improve predictive capabilities, aiding informed decision-making by port authorities. By integrating these approaches, the dissertation advances data-driven solutions for optimizing commercial waterway operations and infrastructure investments. The first stochastic optimization model presented minimizes the sum of port infrastructure investment and expected supply chain cost, considering uncertainty in commodity demand. Applied to the Arkansas section of the McClellan-Kerr Arkansas River Navigation System (MKARNS), it reveals potential annual costs of up to $21 million without accounting for demand uncertainty. To address the limitation of excluding port disruption scenarios, an extended model incorporates port closure uncertainty. This model, utilizing reinforcement learning to enhance efficiency, outperforms traditional methods for large-scale problems. Additionally, machine learning models are developed to overcome data latency issues inherent in traditional datasets like Waterborne Commerce Statistics (WCS). Leveraging Automatic Identification System (AIS) data and historical WCS data, the Long Short-Term Memory (LSTM) model achieves promising accuracy in predicting commodity volume at port terminals. The methodologies developed in this study propel the advancement of waterway transportation systems by optimizing cost and addressing data latency. By integrating stochastic optimization and machine learning techniques, these methodologies enable informed decision-making, reduce costs, and enhance data-driven investment. Their implementation supports sustainable economic development by improving resource utilization, minimizing environmental impact, and fostering regional prosperity.

Download

Share

COinS