An Algorithm for Multistage Dynamic Networks with Random Arc Capacities, with an Application to Dynamic Fleet Management

We consider the class of multistage dynamic networks with random arc capacities a framework that is well suited to model dynamic fleet management problems. We propose a successive convex approximation approach that produces an approximation to the expected recourse function which captures the future effects of current decisions under uncertainty. This method decomposes the network in each stage into tree subproblems, whose expected recourse functions are easy to obtain. We also compare this method with two alternative methods on a set of dynamic fleet management problems. The numerical results show that this method is superior to the two alternative methods.