Ingénierie et gestion industrielles

Forecasting System Monitoring under Non-normal Input Noise Distributions

Abstract

Hoda Sabeti, Omar Al-Shebeeb and Majid Jaridi

In quantitative forecasting models and tracking signal methods, input noise is often assumed to be normally and independently distributed. The goal of this research was to study the distribution of tracking signal and build new monitoring schemes for when the input noise distribution is not necessarily normal. A demand process in the Wilson inventory model was simulated using several input noise distributions. The effectiveness of a proposed tracking signal model was evaluated and compared to existing methods using an inventory cost model. It was found that it is not realistic to assume a normal distribution for the tracking signal even when the noise is normal. Because of the dependency of tracking signal elements, and since there is no specific distribution for it, we used simulation to estimate the best value for the standard deviation and suggest ±3 íÂ�?Â€ íÂ�?Â€íÂ�?Â€ as the control limits. We compared this value with those suggested by other papers, and showed that the proposed limits work better when the process is under control and also when there are different amounts of shifts in mean demand. We also studied different values for the tracking signal smoothing parameter and analyzed the inventory costs for each.