Comparison of Automatic Monitoring Systems in Automatic Forecasting

摘要

The paper compares the performance of several monitoring methods in terms of their ability to detect particular process disturbance types. The forecast errors are either normally and independently distributed, or are the autocorrelated errors that result from the application of the Holt-Winters extrapolation method to disturbed trendwise time series, or from the application of single exponential smoothing to disturbed time series randomly fluctuating around a constant level. In contrast to other research in the field, aside from the step change, also changes in trend and variance are considered. The main conclusion states the cusum method as the best choice in almost all simulated situations. As such the cusum method is recommended as its superiority more than compensates for the extra effort involved.