A Data-Driven Aggregate Production-Distribution System for Perishable Product Retailers
DescriptionPerishable goods are typically ordered, produced, and delivered on a daily basis. Due to the high demand volatility and short shelf lifetime, retailers have been facing the challenges to match perishable supply with uncertain demand. Prior literature has been focusing on inventory policy design and clearance price optimization to reduce sales loss due to demand misestimation. However, it is still not clear how to reduce the inevitable demand uncertainties at the individual store level. Furthermore, most of the prior studies on perishable inventory planning are built on the pre-defined demand distribution. However, in real-world implementation, the parameter distribution is sensitive to some external factors which change dynamically.To address the aforementioned issues, this proposal aims to develop a data-driven inventory planning system for perishable goods by integrating data mining techniques and robust inventory optimization algorithms. Specifically, we first propose an aggregate demand forecasting algorithm to guide factories for an L-day ahead production plan. The aggregate production is conducted at the station cluster level based on a deep sequential model, which reduces the demand volatility while preserving demand sensitivities to external factors. The prediction result, together with the prediction error distribution, is then used as parameter inputs for factory production optimization under certain constraints and inventory policies. The production amount, which is delivered after L days, is then distributed to retail stores based on a next-day demand forecasting and the day-end inventory at the individual store level. The proposed data-driven "Lday ahead aggregate production and next-day redistribution'' inventory planning attempts to provide a new scheme for perishable inventory control with the support of data mining and deep learning algorithms. A field experiment will be conducted to test the effectiveness of our designed system using real-world supply chain data and feedback from potential industry practitioners.
|Effective start/end date||1/01/21 → …|