Research On Application Of The FNN Inverse Soft-sensing Method In The Penicillin Fermentation Process

Penicillin is the first antibiotic which is purified at large scale for clinical, and is the biggest demand antibiotic of various countries. Because the penicillin is the secondary metabolite of the penicillin bacteria, the mechanism of the fermentative process is extremely complex, nonlinear and uncertain. Due to the lack of bio-sensors, some of key biological parameters are so difficult to measure on-line. It makes hard to achieve optimal control of fermentation process. Developmental soft-sensor technique provides an effective way to solve this problem.Taken the fermentation process of penicillin as example, on-line measurement of key biological parameters and project implementation are researched. When the inverse system of penicillin fermentation process is established, the soft-measuring method of fuzzy neural network inverse is applied to measure on-line. And the design and implementation of the software of soft-measuring method is explained.(1)Based on the non-structural dynamic model of penicillin fermentation process, the subsystem "assumed inherent sensor" is constructed, and the reversibility of the system is testified. Then fuzzy neural network and a number of differentiators constitute the soft-sensor model of penicillin based on FNN inverse system.(2)Three separate fuzzy neural networks are adopted to approximate the nonlinear functions of inverse system. Clustering algorithm is introduced to fuzzy neural network based on gradient descent learning algorithm to avoid excessive number of rules. The network structure is relatively simple, which better reflect the characteristics of input and output samples. The design improves the training speed and generalization ability of neural networks.(3) The software of the FNN inverse soft-sensing system used for the penicillin fermentation process is programmed based on the modularization theory. User interface is created by VC + +, which call the MATLAB soft-measuring algorithm, and the database is supported by ACCESS. The software is carried out by a general design, and the specific function of each sub-module is implemented.The results show that the system could forecast the value of the key biological parameters in fermentation process with a high accuracy, which basically achieves the desired goal and meets the demand of design, and provide the basis for advanced control of fermentation process.