Parameter estimation in Crystal Sugar production With MLR, ANN and ANFIS

Abstract

The sugar production process is a complex process in which many variables interact. The cost and time requirements of complex processes are reduced by computer-based modeling techniques and necessary actions can be taken regarding the obtained product quality. In this study for the crystallization stage, solution color which is one of the quality control criteria for sugar production, was predicted by multiple linear regression (MLR), artificial neural network (ANN) and adaptive neural fuzzy inference system (ANFIS). Production data (brix, purity, pol, pH, ash, color and vacuum temperature) obtained from Ankara Sugar Factory General Directorate. As a result of the sensitivity analysis ash, color and vacuum temperature was determined to be the most effective parameters on the estimated output and used as a model input variables. R and MSE values were used as model performance criteria. ANFIS showed better prediction performance than MLR and ANN, R= 0.99.