Neural network and hybrid model: a discussion about different modeling techniques to predict pulping degree with industrial data

Three models to predict kappa number in a pulp mill have been compared. The deterministic model showed expected behavior and was later used in the hybrid model. The pure network model was able to reproduce mill values with satisfactory accuracy, after network optimization and training set filtering. With the introduction of theoretical knowledge in the network structure, the hybrid model results demonstrated a better prediction efficiency and reduced training time.