Date of Graduation


Document Type


Degree Name

Master of Science in Computer Science (MS)

Degree Level



Computer Science & Computer Engineering


Wing-Ning Li

Committee Member

Gordon Beavers

Second Committee Member

Craig W. Thompson


Social sciences, Applied sciences, Algorithms, Transportation


This thesis investigates a logistics problem facing companies that export their products to other countries. The problem is called export intermodal transportation problem. In the export intermodal transportation problem, goods ordered by overseas customers need to be transported from production plants or warehouses of an export company to the customers destinations overseas. The transportation involves using multiple transportation modes such as trucks and rails for the inland portion and ocean liners for the overseas portion, and its objective is to have the goods moved and the cost minimized subject to various constraints. Cost can be minimized by combining orders from different customers to reduce the number of trucks, rails, or ocean containers used, and by selecting the appropriate transportation modes, routes and carriers.

This study provides a formulation of the export intermodal transport problem and proposes two approaches to solve a relaxed version of the problem, where the time constraints are ignored. The first approach divides the problem into three sub-problems: order consolidation on ocean container, ocean port and carrier selection, and inland transportation mode and carrier selection. Order consolidation on ocean container is formulated as the bin packing problem and is solved by the first-fit decreasing algorithm. Ocean port and carrier selection is formulated as minimum cost maximum flow and prototyped with the cycle cancelling algorithm. And finally inland transportation mode and carrier selection is formulated as variable sized bin packing with costs and is solved by a proposed heuristics algorithm. The second approach is a backtracking approach aimed at getting the optimal solution for smaller problem instances and establishing a baseline to compare solutions obtained by the first approach.

Both approaches are implemented as prototypes and evaluated with historical real world data provided by a large food export company. For all data sets, both prototypes produce

solutions with transportation cost less than that obtained by the company manually. On average the prototypes reduce the cost by 3% and save $30,000 for each data set. The three stage solution approach prototype runs much faster than the backtracking approach prototype. For almost all larger data sets, it takes too long for the backtracking prototype to complete. If we let the backtracking prototype run for 30 minutes and keep the best solution, the solutions obtained by both prototypes are comparable in terms of their cost. As for time, the three stage solution approach prototype takes about 2 seconds to obtain each solution.