摘要: |
We consider dynamic vehicle routing under milk-run tours with time windows in congested transportation networks for just-in-time (JIT) production. The arc travel times are considered stochastic and time-dependent. The problem integrates TSP with dynamic routing to find a static yet robust recurring tour of a given set of sites (i.e., DC and suppliers) while dynamically routing the vehicle between site visits. The static tour is motivated by the fact that tours cannot be changed on a regular basis (e.g., daily or even weekly) for milk-run pickup and delivery in routine JIT production. We allow network arcs to experience recurrent congestion, leading to stochastic and time-dependent travel times and requiring dynamic routing decisions. While the tour cannot be changed, we dynamically route the vehicle between pair of sites using real-time traffic information (e.g. speeds) from Intelligent Transportation System (ITS) sources to improve delivery performance. Traffic dynamics for individual arcs are modeled with congestion states and state transitions based on time-dependent Markov chains. Based on vehicle location, time of day, and current and projected network congestion states, we generate dynamic routing policies for every pair of sites using a stochastic dynamic programming formulation. The dynamic routing policies are then simulated to find travel time distributions for each pair of sites. These time-dependent stochastic travel time distributions are used to build the robust recurring tour using an efficient stochastic forward dynamic programming formulation. |