Restricted Recourse Strategies for Dynamic Networks with Random Arc Capacities

We consider a class of multistage stochastic programming problems that can be formulated as networks with random arc capacities. Large problems have proved intractable using exact methods and hence various approximations have been proposed, ranging from approximating the recourse function to sampling a small number of scenarios to capture future uncertainties. We explore the use of specialized recourse strategies that are not as general as network recourse but nonetheless capture some of the important tradeoffs. These new recourse strategies allow us to develop approximations to the recourse function that can be used to solve problems with thousands of random variables. Given these approximations, classical optimization methods can be used. The concept of hierarchical recourse is introduced and used to synthesize and generalize earlier notions of nodal recourse and cyclic recourse.