Key points are not available for this paper at this time.
We consider a stochastic version of a dynamic resource allocation problem. In this setting, reusable resources must be assigned to tasks that arise randomly over time. We solve the problem using an adaptive dynamic programming algorithm that uses nonlinear functional approximations that give the value of resources in the future. Our functional approximations are piecewise linear and naturally provide integer solutions. We show that the approximations provide near-optimal solutions to deterministic problems and solutions that significantly outperform deterministic rolling-horizon methods on stochastic problems.
Godfrey et al. (Fri,) studied this question.