Predicting kappa number in a kraft pulp continuous digester: A comparison of forecasting methods

Abstract

This paper discusses kappa number prediction models using Single Exponential Smoothing, Multiple Linear Regression Analysis, the Time Series Method of Box-Jenkins (ARIMA) and Artificial Neural Networks. Applying a database of an industrial eucalyptus Kraft pulp continuous digester, these four different methods were evaluated according to different statistical decision criteria. After fitting the parameters of the models, validations were performed using a new dataset. Results, advantages and limitations of the four methods were compared. Some statistical forecasting indexes indicate that the ARIMA model showed more accurate estimation results, achieving a MAPE lower than 3 % and over 90% of the prediction data deviations lower than one kappa unit.