Research On Soft-Sensing Methods For Estimating Biochemical Parameters In Nosipeptide Fermentation Process

Fermentation is a common mode of production in modern process industry, which has been widely used in the production of medicine, food, chemical products, and so on. During fermentation processes, the real-time access of some important biochemical parameters such as biomass and substrate concentration is of great significance for process optimization and control. However, until now there are no practical on-line sensors that can be used to measure them directly. In the industrial production, biochemical parameters are mostly measured through sampling analysis. Off-line analysis has large time-delay, which brings many difficulties for the real-time monitoring of important biochemical parameters and the direct quality control of this process. Therefore, for fermentation processes, one of the key problems to be solved is how to fulfill the on-line measurements of important biochemical parameters under existing technical conditions.Soft-sensing is a technique that estimates the hard-to-measure variables using easily available process variables, which provides an effective way for solving the above problems. Taking Nosiheptide fermentation process as research background, in combination with the mechanism of Nosiheptide fermentation process, deep research and discussion on the soft-sensing modeling methods and related problems are performed for the key biochemical parameters that are hard-to-measure on-line in Nosiheptide fermentation process.Based on deep analysis of the mechanism of Nosiheptide fermentation process, multistage characteristic of Nosiheptide fermentation process is mainly studied, and two biochemical parameter soft-sensing modeling methods based on multistage characteristic are proposed in this dissertation. Meanwhile, thorough research and discussion on problems related to soft-sensing technique are performed.(1) Discussion and research on RBF neural network soft-sensing modeling method for Nosiheptide fermentation process are performed, and improvements for several related key problems are provided. Aiming at the problem of secondary variable selection, a secondary variable selection method based on the implicit function existence theorem is proposed; Aiming at the problem of outlier identification, a k-NN algorithm based outlier identification method is proposed; By taking into account the fact that the state trajectories of fermentation processes are closely dependent on fermentation initial conditions, a weighted RBF neural network based soft-sensing identification modeling method is presented.(2) In the case that sufficient input and output data of Nosiheptide fermentation process can be obtained, a phase-identifying-oriented black-box model based biochemical parameter soft-sensing method is proposed for Nosiheptide fermentation process. For multistage characteristic of Nosiheptide fermentation process, a phase identification method based on FCM algorithm and neural network technique is presented. And based on the weighted RBF neural network soft-sensing modeling method, a phase-identifying-oriented black-box soft-sensing model for biochemical parameters is constructed.(3) In the case that no sufficient input and output data of Nosiheptide fermentation process can be obtained, a phase-identifying-oriented series hybrid model based biochemical parameter soft-sensing method is presented on the basis of simplified Nosiheptide fermentation process mechanism model obtained under some assumed conditions. Meanwhile the improved training algorithm for series hybrid model is presented, and the prediction of unknown parameters is realized by using the weighted RBF neural network modeling method.(4) Based on the existing computer monitoring and control system for Nosiheptide fermentation process, the implementation scheme of biochemical parameter soft-sensing system is presented for Nosiheptide fermentation process, and the role of some related function modules is also expounded. Meanwhile, the soft-sensing model updating method is analyzed and discussed. Finally, the soft-sensing methods proposed in this dissertation are applied to identify the draw-off timing of Nosiheptide fermentation process.The above mentioned methods provide an effective way for the implementation of the real-time monitoring of important biochemical parameters, and also lay a solid foundation for the implementation of direct quality. In the case that no sufficient modeling data can be obtained, by using series hybrid model based soft-sensing method which combines mechanism model and data, the unknown parameters of mechanism model are identified with little modeling data, and then the real-time prediction of key parameters is realized. In case that sufficient process input and output data can be obtained, to avoid the defect that the simplified mechanism model couldn’t completely describe process characteristics, taking advantage of the characteristic of neural network that it can approximate nonlinear functions with arbitrary precision, the real-time estimation of the key parameters is realized using the black-box model based soft-sensing method. The above methods have been successfully used in Nosiheptide fermentation process, which validates their effectiveness. The methods can be also generalized to other fermentation processes or other similar complex industrial processes.