Soft-sensing modeling method based on Continuous Hidden Markov Model for microbial fermentation process

Considering the temporality of microbial fermentation process, a soft-sensing modeling method based on Continuous Hidden Markov Model (CHMM) for microbial fermentation process is proposed. Firstly, in order to improve the robustness of CHMM, multi-observation training sample sequences are used to train the CHMM. And the modified Baum-Welch parameters re-estimation formula is used to optimize the parameters of CHMM. Then, the new observation vector is inputed to the CHMM model library and the emission probability of each CHMM in the model library is calculated using the Viterbi Algorithm. Finally, the soft-sensing result can be obtained by computing the weighted average. The model is applied to an erythromycin fermentation process, and case studies show that the new approach has better performance compared to the conventional method based on ANN.

1 INTRODUCTION

Microbial fermentation process is a very complex, uncertain and non-linear dynamic process. One of the critical quality indexes is biomass concentration. Until now this variable is mainly determined by off-line laboratory analysis. The development of biologic sensor technology is restricted mainly by the high costs and technical complexity, which makes the on-line measurement of biomass concentration become rarely using in common industrial bioprocess control systems [1]. Due to these limitations, a soft-sensing technique which can estimate immeasurable variables has been proposed for fermentation systems [2]–[3].