Industrial Forecasting Using Knowledge-Based Techniques and Artificial Neural Networks

Abstract

Most techniques used for forecasting can be classified in under three main categories. The first treats the demand as a time series and predicts it using different time series analysis techniques [1], [2]. In the time series forecasting approach several methods (like, for example, the exponentially smoothing technique) have been developed to give different weight to more recent data, to monitor the forecasting error and to adjust the smoothing factor to the evolution of the error using Trigg’s warning signal [2]. General problems with the time series approach include the inaccuracy of prediction and numerical instability due to lack of information about external factors that influence the model and which are not included in the time series historical data [3], [4]. The second approach is based on regression techniques which accept that demand is heavily dependent upon external factors.