Forecasting intermittent demand in large scale inventory system.

摘要

This document summarizes the findings from the author's doctoral research in forecasting and in developing forecasting methods for large scale inventory systems with intermittent demands. Intermittent demand is characterized by demand data that has many time periods with zero demands. It is hard to model intermittent demand by conventional distributions. It is commonly found in military supply network among reparable spare parts. The availability of spare parts, which depends on the accuracy of forecast, affects the accomplishment of mission. Intermittent demand is also found when forecasting is done at the stock keeping unit level. The goal of this research is to improve demand forecasting and inventory control in large scale inventory systems with intermittent demands. This is achieved by a four prong approach. One, this research developed an individual intermittent demand forecasting technique named MCARTA with performance better than that of other relevant intermittent demand forecasting techniques and robust across hard to forecast demand scenarios. Two, a new metric to measure the performance of a forecasting technique within the context of an inventory system, named the FLTDD divergence metric is developed via the lead time demand distribution. Third, this research identified error measures with which behavior of operational performance measures can be predicted. Based on these two applications the effect of forecasting techniques in an inventory system can be evaluated. Finally, this research developed a meta-forecaster by applying supervised statistical learning, with which the selection of the best forecasting techniques for an item can be made efficient in a large scale inventory system.