An on-line adaptive control strategy based on DO/pH measurements and artificial neural network pattern recognition (ANNPR) model for fed-batch cultivation processes was proposed. Various changing patterns of pH and dissolved oxygen concentration (DO) under pH-Stat, DO-Stat, and the conditions of substrate in excess were collected and used to train the ANNPR models. To further improve the universal abilities of the ANNPR models, extra white noises with appropriate variance were added on the basis of the original training data. Based on the on-line measured pH and DO data, the recognition results on current physiological state was deduced by the ANNPR models, and then the on-line adaptive control of nutrient feeding rate was implemented. Compared with the traditional pH-Stat control, the proposed control strategy increased E.coli cell productivity from 36gDCW/L to 56.7gDCW/L and Pichia Pastoris cell productivity about 123% without byproduct accumulation. Only the online measurement of the most traditional variables of pH and DO were required for the ANN-PR-Control proposed in this paper, and the subsequent control could be easily implemented using very simple regulating rules. The ANN-PR-Control system could assure cell growth at a higher rate and acetate accumulation at very low level at the same time. The cheap, universal, robust, and simple ANN-PR-Control system could be applied for high cell density cultivation of recombinant microorganisms to efficiently express value-added foreign proteins or enzyme in industrialized scale.
|