Study On Soft-sensor Modeling For Predicting Inulinase Concentration At Recombinant Pichia Pastoris Cultivation Process

Fructooligosaccharides has been widely used in dietary supplements because it has many advantages, such as indigestibility, caries resistance, improving lipid metabolism and so on. Hydrolysis of inulin with endoinulinase that obtained by Pichia pastoris is one of the important ways for producing fructooligosaccharides. Pichia pastoris fermentation process involves many important biochemical variables, which include yeast concentration, methanol concentration and inulinase activity concentration. The yeast concentration and methanol concentration can be measured by corresponding laboratory-scale online measuring instruments. But inulinase concentration still relies on the offline analysis of enzyme activities, which not only consumes lots of manpower and resources, but also affects the implementation of real-time control strategy and improvement of fermentation technique.Soft sensing is one of the effective ways to solve the above measurement problem. To estimate and predict inulinase concentration in Pichia pastoris fermentation process, two soft-sensors, which are based on biological reaction mechanism and support vector machine respectively, are established in this paper.Soft-sensor based on biological reaction mechanism is more suitable for the situation in which the quantity of input variables is not too large. Unstructured mechanism soft-sensor is based on the mass balance equations of reaction components, combining Logistic equation which is used to describe the specific growth rate of Pichia pastoris. It’s simple and universal, but the estimation results are lack of accuracy. In order to improve the estimation accuracy, metabolic mechanism soft-sensor is established, which is based on the material flow balance of Pichia pastoris’metabolic process. By analyzing Pichia pastoris’ methanol metabolic pathway, metabolic model of specific growth rate is derived to replace the Logistic equation. Then, three methods are used to optimize the unknown parameters appearing in the above two soft-sensors, which are Levenberg-Marquardt algorithm(referred to as the L-M algorithm), genetic algorithm and particle swarm optimization algorithm. By Comparison, the estimation accuracy of particle swarm optimization algorithm is proved to be the best. Combining with particle swarm optimization algorithm, two mechanism soft-sensors both can be used to estimate the inulinase concentration and the estimation accuracy of metabolic mechanism soft-sensor is better than the unstructured mechanism soft-sensor.Soft-sensor based on support vector machine is more suitable for the situation in which there are more types but less numbers of input variables. Based on the least square support vector machine regression, soft-sensor is established to predict inulinase concentration during the Pichia pastoris fermentation process. To eliminate the linear correlation of input variables, principle component analysis is used to reduce the input dimension and convert dependent variables into independent variables. In addition, leave one out algorithm and particle swarm optimization algorithm are combined to optimize the soft-sensor’s parameters. Leave one out algorithm is used as the target function to minimize the error of cross validation, and particle swarm optimization algorithm is used to search the best parameters. The experiental results show that the proposed soft-sensor has better prediction accuracy than the soft-sensor based on standard support vector machine. In addition, both the principle component analysis method and the proposed parameter optimization method can improve the prediction accuracy significantly.To sum up, according to the type of input variables and the amount of samples, soft-sensor based on biological reaction mechanism or support vector machine regression can be used to estimate and predict the inulinase concentration during the Pichia pastoris fermentation process. It can provide references and guidances for the implementation of real-time control strategy and improvement of fermentation technique.