Study Of Soft Sensing Method Based On Continuous Hidden Markov Model In Fermentation Process

Biological parameters are the important variables in the fermentation process, which reflect the state of the fermentation process and impact the quality and yield of the output. In order to control the fermentation process effectively, online measurement of some key biological parameters is indispensable. However, in the actual fermentation process, due to technical and technological constraints, on-line measurement of some key biological parameters is difficult. Off-line analysis by manual sampling method is used commonly, which is difficult to meet the real-time monitoring and optimization control requirements of fermentation process. To overcome these problems, a way-out is to build a soft sensor model.This thesis mainly studies on the erythromycin fermentation process. With the mechanism and the soft sensor method, several schemes to solve the problem of the soft sensor are taken out. The Continuous Hidden Markov Model (CHMM) is introduced into the field of soft sensor modeling, and a new method based on CHMM is proposed and applied in the soft sensor modeling of the biomass estimation.The main contributions of the thesis are as follows:1. An overview is made on the soft sensor modeling method and the basic theory and algorithm of CHMM, CHMM based soft sensor modeling method is studied for the erythromycin fermentation process.2. A soft sensing credibility evaluation index is proposed to avoid blindness problem during the practical application of soft sensing result to monitoring in fermentation process.3. To solve the defects of traditional Baum-Welch algorithm and the practical problems of CHMM in soft sensor modeling applications, modified Baum-Welch parameters revaluation formula is proposed. And the CHMM fermentation steps of the soft sensor modeling based on modified Baum-Welch algorithm are given.4. In order to improve the prediction accuracy of CHMM-based soft sensor, the distinctive training algorithm is used instead of the traditional training algorithm. The training algorithm based on minimum classification error criterion is proposed, and the CHMM fermentation steps of the soft sensor modeling based on improved MCE criteria are given.5. With the process data, soft sensor models of CHMM based on modified Baum-Welch algorithm and improved MCE criteria are established for biomass estimation in the erythromycin fermentation process. The testing result shows the effectiveness of the proposed methods and the practical significance of the credibility evaluation index.