In the pulp and paper and biofuel industries, real-time online characterization of biomass gross calorific value (GCV) is of critical importance to determine its quality and price and for process optimization. Near-infrared (NIR) spectroscopy is a relatively low-cost technology that could potentially be used for such an application. However, the NIR spectra are also influenced by biomass temperature (T°) and moisture content (MC). In this paper, external parameter orthogonalization (EPO) is employed to remove simultaneously the influence of T° and MC on the spectra before predicting GCV. EPO is of particular interest when one desires to transfer information from one modeling experiment to another, such as when developing a calibration model for a new property from the same material, or when it would be more efficient to divide the experimental effort. EPO was found to be an effective method for desensitizing a PLS calibration model to the influence of T° and MC, enabling robust and accurate prediction biomass GCV. Partial least squares (PLS) regression models developed with EPO always provided equal or better performance than models developed without EPO. The paper shows that experimental efforts and costs can be reduced by approximately one half while maintaining prediction accuracy and model robustness.
Ajoutez cet article à votre liste de sélections pour demander le PDF - Add this item to your selection list to request the PDF