In this dissertation, we present and analyze three models of sequential decisionmaking that are motivated by information issues in the trucking industry and more broadly in transportation. The first model relates to the common need to determine a path along a road network in which the arc costs are known precisely. We develop an algorithm suitable for use by a centralized information service providing route guidance to dispatchers or drivers. The algorithm exhibits a form of machine learning, so that fewer computer resources are needed to determine optimal paths once. the system has been trained. The algorithm operates approximately 60% more efficiently than the best-known alternative algorithm for this problem.; The second issue addresses the question: what is the economic value of near real-time traffic data. Such data can lead to re-routing enroute around traffic congestion. Such data may be corruted by noise and available only after a delay due to data collection, fusion, and transmission. We introduce the concept of the time value of data, to help address this question.; In the third problem we assume that the decision-maker has access to historical distributions of link travel times at different times of the day. The are costs are modeled as random variables whose distributions are nonstationary. Rather than a path, from source node to destination node. the optimal solution is a time-dependent routing policy. We demonstrate that a heuristic search algorithm quickly identifies a relevant subspace of the state space, and produces the optimal value function and an optimal policy for that subspace, and can significantly out-perform a standard application of dynamic programming.
展开▼
