Learning Inventory Control Rules for Perishable Items by Simulation-Based Optimization

Abstract

We consider an inventory control problem for quickly perishable items, such as fresh produce, at the retail store level, assuming a fixed shelf life. Demand is affected by both uncertainty and seasonality within the week, as sales feature a peak close to weekends. Another complicating factor is customer behavior and inventory issuing: In the case of a first-in-first-out (FIFO) pattern, older items are sold first, whereas a last-in-first-out (LIFO) pattern is more critical as newer items are sold first, which may increase scrapping. These and possibly other complicating factors make elegant mathematical modeling not quite feasible. Hence, we experiment with simulation-based optimization approaches integrating a discrete-time simulation model with direct search methods like simplex and pattern search. The approach is rather flexible, and learning simple rules has a definite advantage in terms of management acceptance. One aim is to compare simple order-up-to rules, based on overall available inventory, against more complex rules that take inventory age into account. Since more complex rules require more effort in implementation, it is important to understand under which circumstances their use is justified. We also want to study the effect of economic parameters, demand uncertainty and skewness, as well as FIFO/LIFO behavior. Preliminary computational experiments are reported, including a comparison with simple newsvendor-based heuristics.