Abstract:
Inventory management has been researched for many years in the literature. The problem of determining the optimal inventory level directly affects the profitability of the systems. This thesis proposes an optimization framework for multi-item, multi-location inventory management systems to obtain optimal inventory levels for multiple periods. The distribution of demand is unknown, it is estimated based on the probabilities generated by a random forest. Backlog is not allowed, demand is lost in case of stock-out. The first analyses conclude that as the scenario number considered in the mathematical model increases, the system stabilizes and performs better due to the consideration of variation in demand. To evaluate the robustness of the model, the suggested linear programming model is tested on different inventory level cases and it is concluded that inventory cases have important effect on objective values of the system. The analyses are constructed on two systems with different management strategies: firstly, the system is considered to allow lateral transshipment along with the replenishment, then the system is assumed to allow only replenishment decisions. It is concluded that allowing transshipment decisions improve the lost sales of the system. After the case analyses, objective value distribution for scenarios are analyzed to evaluate the robustness of the proposed method. Finally, the periodic review strategy is tested and it is concluded that there is a significant difference between the systems with periodic review and single review strategy. In addition to the prediction methods used in the case analyses, smart MA is introduced and the analyses show that it is a powerful prediction method for the simulated system.
