Application of Artificial Neural Network and SARIMA in Portland Cement Supply Chain to Forecast Demand

Supply chain management (SCM) is currently a hot issue of discussion, though the first step of SCM is how to adjust units in order to forecast demand accurately for the future. The cement demand has significance in seasonality and trends. In general, the cement demand in developing countries is higher, while the cement demand in developed countries diminishes to a steady level. For past twenty years, Taiwan has experienced a similar path. This research focuses on the cement demand in Taiwan for past twenty years, which conduct data collection and relation analysis. Furthermore, it establishes quarterly and monthly cement forecast model. The two applied methods are seasonal ARIMA and artificial neural network (ANN). By comparing the demand data from January 2004 to March 2005, it verifies the accuracy of each forecast model. From the research result, the established forecast model from ANN presents a most accurate outcome of averaging value within 3%. Therefore, this research suggests that applying ANN with quarterly unit to forecast is the most accurate model. Due to cement is highly influenced by weather and Chinese new year festival period, the monthly unit is not appropriate and would cause significant deviation and difficult to process by mathematics or statistic formula. Applying quarterly unit has shown a stable condition during data presentation. Although during verification process that some points have shown zero error condition, but it is recognized as a trustable forecasting method in cement demand forecasting in Taiwan.