Soft Sensor Model For Penicillin Fermentation Process Based On SVR Optimized By Adaptive Quantum Particle Swarm Optimization Algorithm

Penicillin is a kind of the most common antibiotics in clinical application, which plays an importantrole in the medical and health field and has a significant influence to our daily life and industrial production.The penicillin fermentation process controlled by many factors is extremely complex. In order to improvethe product yield and quality of products, it needs to acquire the real-time parameter information during thereaction process to adjust and control the fermentation process, so that enterprises can get the maximumbenefit. However, some important biological variables in the process such as penicillin concentration,biomass concentration and substrate concentration, are difficult to measure on-line with biosensortechnology, which becoming a bottleneck problem of control optimization of the penicillin productionprocess and seriously affecting the automated production of penicillin fermentation process. Therefore, inorder to improve the product yield and quality, it is quite important for the penicillin fermentationindustrialization to establish an accurate prediction model to control and optimize the fermentation process.The research topic in this paper is sponsored by National Natural Science Foundation of China namedas “the Research on Adaptive Modeling and Multi-objective Collaborative Optimization Control ofBiochemical Process Based on Data”, which mainly focuses on soft sensor model based on SVR (SupportVector Regression) optimized by AQPSO (Adaptive Quantum Particle Swarm Optimization). The primarycontent of the study includes the following four aspects:(1) To better control and optimize the penicillin fermentation process, firstly, we have in depthanalysis to the factors that influence its fermentation process, and then the soft sensor modeling method isproposed to solve the problem that some parameters are difficult to measure on-line, which lay a theoreticalfoundation for following research.(2) Penicillin concentration, biomass concentration and substrate concentration are separatelypredicted by the soft sensor model used SVR algorithm of fermentation process. It can be seen from theexperiments that change of parameter affects the model performance. Parameters which depend on thesubjective experience cannot meet the high precision prediction requirement of industrial production, so theprediction precision and training time need to be improved. (3) For the problem that the parameters of SVR model are uncertain this article adopts AQPSO(Adaptive Quantum Particle Swarm Optimization) algorithm in the selection of regression parameters forSVR to build online prediction model of penicillin fermentation process. This model has faster convergencespeed and solves the defect of prematurity, owning stronger global optimization ability and ability oflearning and generalization.(4) Input and output variables of AQPSO-SVR model are determined by analyzing the fermentationprocess, and then validate the performance of the model on the penicillin control system of the laboratory.Through the experiments, it is found that this model which reduces runtime greatly has higher predictionprecision compared with other algorithms, perfectly realizing the accurate prediction to some importantbiological parameters and validating the feasibility and validity of the algorithm in the modeling of softsensor for penicillin fermentation process.