Inventory optimization and optimal distribution policies in substitutable product distribution networks

In multi-product inventory systems, considerably savings in specially transportation costs may be achieved by coordination of replenishment orders for groups of products. Moreover, product substitutions in these systems reduce significantly inventory costs.In this thesis, a fully substitutable multi product system is considered in which customer demands may be satisfied by delivering any combination of products. In the studied system, product substitution is driven by firm and product inventories may be replenished by coordinated control. The aim of the thesis is to determine order fulfillment and replenishment policies minimizing approximately average cost in unit time in infinite horizon.In this thesis, a specific control system is considered which is known as a (S,c,s) type policy, and works in the literature are used to determine the control parameters. The order fulfillment problem is modeled as a semi Markov decision process, and reinforcement learning algorithms with approximation to Q factor function by artificial neural network are used to solve this problem. In the thesis, an algorithm is developed which determines iteratively order fulfillment and (S,c,s) policies for each product in a multi-product inventory system. In numerical examples, it is seen that order fulfillment and replenishment policies determined by the developed algorithm outperform myopic policies.