Improvement And Implementation Of The ANN Inverse Soft-sensing Method In The Erythromycin Fermentation Process

To deal with the low precise measurement of the"Assumed Inherent Sensor"based ANN(artificial neural network) inverse soft-sensing method in the erythromycin fermentation process, several methods of improvements are presented to increase the precision of soft-sensor and applied to the erythromycin fermentation process in a PHysic Co.. The good result approves the methods are effective, and can increase the estimation precision of biologic control variables, including mycelium concentration, sugar concentration and production concentration, online in the erythromycin fermentation process. The main research contents are as follows.1. On the foundation of the"Assumed Inherent Sensor"based ANN inverse soft-sensing method, several important secondary variables are added according to the extending frame of ANN inverse system and the erythromycin fermentation status, then the secondary variables are disposed with the PCA (principal component analysis). Thus the soft-sensing precision is advanced greatly.2. Some secondary variables of the ANN inverse soft-sensing method are first pretreated by the Two-Step-Judgement method and then reprocessed by the moving average method. Thus the noise can be filtered effectively, and the gross error can be prevented and revised properly.3. A new normalization method on the variables is implemented to make the data embody more characteristics of the variables.4. As the module of training and testing neual network is appended, as well as the module of storaging and interpolating the off-line assayed data, and the module of selecting in the listed versions of the ANN inverse soft-sensor, the online revision of the soft-sensing model is implemented. Thus the selected soft-sensing model can guarantee the soft-sensing precision.5. The software of the ANN inverse soft-sensing method used for the erythromycin fermentation process is programmed based on the modularization theory.