Anticipatory Vehicle Routing by means of Optimal Learning

• Applicant: Dr. Stephan Meisel
• Subject Area: Accounting and Finance
• Term: from 2012 to 2013
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 214711072

A large number of business related vehicle routing problems are subject to increasingly strong stochastic influences. Due to trends such as electronic commerce, mass customization and the increasing focus on service quality, logistics service providers continuously receive new information that is relevant for planning of vehicle routing operations. Technological innovations such as GPS and mobile communication enable the use of information updates for modification of existing routes while vehicles already are on their way. By repeatedly executing route modifications, service providers aim at more efficient use of vehicles as well as on an increase of customer satisfaction.The resulting planning problems must be considered as multistage decision problems subject to uncertainty. Finding a solution to these problems requires anticipatory optimization procedures, which generate high quality routes by implementing an a priori reaction to the expected remaining decision process as soon as new information arrives.An optimal decision realizing perfect anticipation may be derived by stochastic dynamic programming. However, due to the tremendous computational requirements involved, such an approach cannot be applied to a routing problem of business relevant dimensionality.As a consequence, simulation-based procedures such as approximate dynamic programming are considered in order to realize the best possible degree of anticipation. Nevertheless, these procedures still require quite a lot of computational effort and can therefore be applied to only a limited scope of problems up to now. The present research proposal is about integration of techniques from the field of Optimal Learning into simulation-based procedures for anticipatory vehicle routing. The purpose of integration is fundamental improvement of these procedures, reduction of their computational requirements and, thereby, full realization of their potential for anticipatory vehicle routing.

• DFG Programme: Research Fellowships
• International Connection: USA
• Host: Professor Warren B. Powell, Ph.D.