Decomposition in Stochastic Dynamic Programming for Production and Inventory Control

Abstract

The multi-dimensionality of most practical problems is probably the main cause of the lack of applications of dynamic programming. In production-inventory problems for instance one uses rolling plans in general. By choosing the planning horizon large enough one may hope that the decisions made now are not too bad for the future. Instead of the explicit state evaluation of dynamic programming one has here a kind of implicit state evaluation. Exact state evaluation is indeed not possible by the dimensionality of the problems. But heuristic methods of state evaluation may be a good alternative and have been investigated only incidentally. We will illustrate here such a heuristic.