City logistics: Predictive analytics for real-time freight management.

摘要

The challenges of contemporary freight management are moving beyond cost efficiency towards superior customer service, agility, and timely service. By its nature, freight distribution is a stochastic and dynamic optimization problem. It deals with future events in an environment with significant sources of uncertainty. Ignoring the unexpected events during operation may lead to delays, higher costs and inferior customer service. To handle the inherent dynamism, real-time information obtained from recent innovative technologies provides promising improvements in the freight management system. The integration of available real-time information and the utilization of dynamic traffic assignment (DTA) models to obtain prevailing and anticipated traffic conditions on the network is still lacking. The principal focus of this research is to devise good and computationally efficient approaches that would enable a commercial vehicle fleet operation manager, or dispatcher, to take advantage of real-time information to dynamically manage available resources to serve the time-sensitive customer requests while recognizing the prevailing and anticipated traffic conditions on road networks. The dissertation begins with introducing the real-time freight management problem and presenting the modeling framework of the system. The existing literature for the vehicle routing problem and related problems is reviewed to provide an overview of the richness of problems that have been investigated within this field. Next the formal mathematical formulation and solution approach for the time-dependent vehicle routing problem with time-windows (TDVRPTW) is provided followed by the extension of the model integrated with a DTA model and simulator. In the context of real-time fleet management, particular interest is given to one-to-many-to-one pickup and delivery services. The dynamic traffic assignment and simulation model (DYNASMART) is embedded in the model framework to provide both real-time and anticipated traffic information. Four real-time policies are introduced to deal with information gradually revealed in the operation. To demonstrate the model effectiveness after addressing the variations of traffic conditions induced by disruptive events with anticipated traffic information, an application to Chicago urban network on a snowy day is executed.