Modeling the removal of hemicellulose from cereal straw at lab-scale using self-organizing maps followed by multiple linear regression

Modeling is widely accepted to be an important tool for developing reactor management and control strategies. A statistical approach based on self-organizing maps (SOMs) followed by multiple linear regression (MLR) was used in this study to model the digestion process of cereal straw. Thus, a black-box modeling strategy was followed in which the reactor was considered as a system that could be described with four input variables (W, the ratio between straw weight and the volume of NaOH solution; C, concentration of NaOH; T, operating temperature inside the reactor; H, contact time) and a single output variable (DE, the amount of degraded hemicellulose). In order to apply a classical MLR analysis, the original database of 45 cases was divided into two groups: a development data set to build the model and an independent test data set. It is important that both data sets have the same statistical properties and SOM were used for this purpose. Subsequently, a classical MLR analysis was carried out. The model included all candidate inputs (W, C, T and H) in the equation proposed to model reactor response. Nevertheless, W appeared to be the most relevant variable to explain changes in the DE, and nearly 50% of the overall variability can be attributed to its influence. The proposed model explained 86% of the overall variability, thus making it possible to adopt decision-making strategies during reactor operation under different operating conditions.