Design of large-scale logistics systems for uncertain environments

Most logistics systems function in an environment with inherent operational uncertainty. It is often difficult to develop tractable models for expected cost minimization, especially when system designs are complex and/or problems are of large scale. This dissertation adapts a continuum approximation methodology developed for deterministic problems for use in the analysis of uncertain logistics systems. Importantly, the methodology yields expected cost approximations that improve with scale, and allows near-optimal configuration of systems.; The proposed methodology is used to analyze two fundamental logistics systems: load-constrained vehicle routing with uncertain customer locations and demands, and deadline vehicle routing with uncertain customer service times. Advances in information and communications technology now allow cost-efficient dynamic coordination of vehicle fleets during operations. Although such designs may generate substantial cost savings over traditional uncoordinated designs through risk pooling, modeling and computational difficulties have prevented their analysis to date.; For load-constrained routing systems, several coordinated designs are proposed and analyzed. Locally-coordinated designs that allow neighboring vehicles to pool capacity and a globally-coordinated strategy in which all vehicles share capacity are presented. Expected distance approximation models are developed for the designs, and used to develop near-optimal configurations. The model for the globally-coordinated design is validated with simulation; test problem results indicate model accuracy of within 5% for near-optimal configurations. To assess the cost savings of this design, approximation results are compared to a lower bound approximation of a traditional uncoordinated design. In configurations with equivalent service levels, the number of additional vehicles required above a deterministic minimum is reduced 29% to 82% for the test problems. Furthermore, in most problems the strategy reduced the number of expected vehicle-miles in addition to a deterministic minimum with savings from 15% to 45%.; In deadline routing systems, vehicles must service customers and return to the depot by a deadline. For the problem with uncertain service times, approximation models are developed for an uncoordinated fixed-zone design and a locally-coordinated design in which demand is reallocated at a fired time point. Test problem results indicate that the number of vehicles required in addition to a deterministic minimum may be reduced 6% to 28% using reallocation.