Author

James M. Park

Date of Award

3-24-2016

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Operational Sciences

First Advisor

Jeffery D. Weir, PhD.

Abstract

The United States military heavily relies on rail freight operations to meet many of its logistical needs during both peacetime and wartime efforts. As the head organization responsible for managing and overseeing all modes of military transportation, United States Transportation Command depends on timely accurate railcar demand forecasts to drive critical decisions on distribution and placement of railcar assets. However, the intermittent nature of railcar demands based on location and commodity make it a challenging task for forecasters. Furthermore, these “lumpy” demands often come without any obvious trends or seasonality. This study explores the utility of both traditional forecasting methods and newer techniques designed specifically for handling intermittent demands. All forecasting parameters for each method are optimized based on specific cost functions. Accuracy metrics are then applied to all forecasts for analysis. The results indicate that for the Department of Defense’s railcar demands, optimizing basic forecasting methods such as Simple Moving Averages and Simple Exponential Smoothing outperform more popular methods for sparse demands such as Croston’s method and its variants. Despite its theoretical superiority, applying Croston’s method to railcar demands was found questionable and consistently produced poor forecasts compared to other methods. Analysis provides valuable insight in future strategies for forecasting intermittent demands.

AFIT Designator

AFIT-ENS-MS-16-M-124

DTIC Accession Number

Pending

Share

COinS