SVM-based prediction of the product formation for industrial 2-keto-L-gulonic acid cultivation

2-keto-L-gulonic acid (2-KGA), a key precursor in the synthesis of L-ascorbic acid, is produced by mixed fermentation of Bacillus megaterium and Gluconobacter oxydans with L-sorbose as substrate. For such mixed cultivation, the mechanistic modelling is difficult because the interactions between the two strains are not well known yet. Therefore, data-driven modelling is studied in this paper. The rolling learning-prediction (RLP) based on support vector machine (SVM) is practiced to predict the product formation. To satisfy the online application demand, pseudo-on-line prediction is carried out using the data from commercial scale 2-KGA cultivation. The prediction approach receives data in sequence and the historical database of the SVM is updated with statistical analysis of the product formation after the termination of a batch. The robustness of the prediction approach is further tested by adding extra noises to the process variables.