Study on online soft sensor method of total sugar content in chlorotetracycline fermentation tank

• Published in: Open Chemistry 2020-01, Vol.18 (1), p.31-38
• Main Authors: Sun, Yu-mei, Han, Xiang, Zhang, Dong-xiang, Sun, Qiao-yan, Chen, Xiang-guang, Yao, Min-pu, Huang, Su-yi, Ma, De-shou, Zhou, Biao
• Format: Article
• Language: English
• Subjects: chlortetracycline fermentation; Gaussian process regression; output recursive wavelet neural network; soft sensor modeling; total sugar content
• ISSN: 2391-5420; 2391-5420
• DOI: 10.1515/chem-2020-0004

In order to solve the problem that the total sugar content of the chlortetracycline fermentation tank can not be automatically detected online, a prediction method which combines the output recursive wavelet neural network and the Gauss process regression is proposed in this paper. A soft sensor model between the measurable parameters (inputs) and the total sugar content (output) of the chlortetracycline fermentation tank was established. The soft sensor model was trained by self updating algorithm. Based on field data, the accuracy and generalization ability of the soft sensor model were analyzed. It is shown that the prediction accuracy of the combined model proposed in this paper is better than that of other single models. The results demonstrate the superiority of the method, and MRE and RMSE are used to evaluate the performance of the soft sensor model. It shows that the prediction precision of the soft sensor model based on ORWNN-GPR combination is relatively high in the long period of fermentation, and is suitable for on-line prediction of the total sugar content of the chlortetracycline fermentation tank. The soft sensor method can effectively reduce the labor intensity of the analysts and saves the production cost for enterprise.