Dr. Kevin Taaffe, Dr. Aurelie Thiele, Narges Hosseini, Rockey Myall, Wennian Li
Forecasting and optimization have traditionally been approached as two sequential components of inventory management: the random demand is first estimated based on historical data, and the forecast is then used as input to the optimization module. This method suffers from several limitations: (i) the best forecasting method might not be known a priori, and may vary depending on the product considered, (ii) the forecasting module does not take into account the over and underage costs of the inventory model, and instead penalizes over and under-predicting demand equally, although backorder costs are typically much higher than holding costs. This provides motivation for devising an operating strategy that builds upon the information revealed over time and remains flexible to integrate forecasting and inventory control in a framework that can be applied to large numbers of products. In preliminary work, the PI and Co-PI have demonstrated the potential of adaptive forecasting, and achieved cost reductions of up to 5% on test problems. To establish the relevance of the Adaptive and data-driven forecasting and inventory control allows companies to recognize changing demand conditions and mitigate the risk of costly demand estimation errors by (i) tracking the performance of forecasting methods available to the manager and (ii) relying on historical data as much as possible without making assumptions on the underlying probability distributions. It has emerged as an important tool to help companies satisfy customer demand in fast-changing environments. This technical report discusses the challenges related to traditional demand management techniques, motivates the relevance of the data-driven approach to address these issues, and provides a high-level description of the solution tools implemented in this project.
The overall project objective of this research is to investigate approaches to forecasting and inventory control that are (i) data-driven, i.e., dynamically integrate the experimental measurements in the decision-making framework, and (ii) adaptive, i.e., exploit information revealed over time, to reduce part stockouts and provide CELDi partners with a framework well-suited to the information available in real-life logistics problems.
This research has provided an integrated framework to forecasting and inventory control that is more tightly connected to the data and adapts seamlessly to changing demand conditions. It has thus contributed to positioning CELDi at the forefront of inventory management research. The overarching goal has been to help CELDi industry partners recognize the impact of forecasting assumptions on their replenishment strategies, and to provide them with a decision tool integrating forecasting and inventory control using cutting-edge demand management techniques.