| Trucking industry, the business of transporting products via trucks, is vital to the health of our economy for the sheer number of people it employs, the value of product it hauls, and the diversity of dispatching models it uses. One of these dispatching models is the spot-market dispatching (SMD). This form of dispatching went through a recent transformation as a result of the industry deregulation in the 80s and the emergence of the internet in the late 90s. The deregulation allowed easier establishment of new trucking companies and their access to the entire market; the internet allowed for spot freight to be posted on the on-line hosting sites where shippers, brokers and truckers can post their service. This change, however, brings with it challenges and opportunities which is the focus of this dissertation.;In this dissertation, we give a broad introduction to the freight spot-market; we identify the challenges and the opportunities. Spot-market dispatching problem is formulated as a dynamic assignment problem, implemented as a Markov Decision Process (MDP), which has its objective as maximizing the operation profit at the end of the dispatching planning horizon. A freight spot-market loads generation platform is created to mimic the dynamics of trucks and loads in such markets. Platform allows generation of data representing different market settings. Approximate dynamic programming methods are proposed to solve the dispatching problem. To address the curse of MDP state-space dimensionality for real-world settings, Neural Networks were used to approximate the value function. We benchmark our methods with local and myopic policies typically used by dispatchers in the business. Load hosting sites offer information about available loads and trucks across the market. We also explore the use of such information into dispatching policies and study its effect on the overall performance. |
