Control of a logistic node via neuro-dynamic programming

In this paper we consider the dynamic model of a logistic node of a transportation network. A necessary and sufficient condition for the existence of a stable feedback policy is given. Optimal dispatching rules can be derived in the case of small instances either by solving an integer programming problem or by considering a dynamic programming approach. Both methods present a combinatorial complexity which makes them not viable in realistic scenarios. For this reason, a neuro-dynamic programming (NDP) approach is introduced and used to refine a heuristic solution of the problem. The solution computed through the NDP approach aligns with the decisions given by a sub-optimal control policy inspired by the literature available for manufacturing systems.