Inventory management via model predictive control based on demand forecast considering price change by particle filter

Abstract

It is important to manage inventory adequately at a retail stores or wholesales. For this reason, we propose an inventory management method based on model predictive control. Model predictive control is an optimal operation method of dynamical systems and the process of stocks is able to be seen as a dynamical system. Therefore, by using model predictive control, there is a possibility that orders can manage their items properly for various evaluation criteria, such as reducing order cost and equalizing the order quantity. In order to achieve order's purpose, we need to improve a long-term prediction accuracy. Consequently, we also propose a stochastic demand forecast model based on a state space model and use a particle filter that one of non-liner and non-Gaussian filters for this model so as to make accurate demand forecasts. In our proposed model, we consider changes in demand due to price changes, such as changes in consumption tax rate, and adopt an idea from Prospect Theory in behavioral economics. In general, it is difficult to identify parameters in the state space model. For this purpose, we use a self-organizing state space model to solve this problem. In the self-organizing state space model, parameters are included in the state vector and they are estimated simultaneously online. We demonstrate that a proper inventory control is achieved by our proposed method using actual sales data of canned beer at a real retail store and show its effectiveness by comparing the conventional method.

Keywords

• invenory management
• model predictive control
• particle filter
• demand forecast
• state-space model
• prospect theory