Hybrid Modeling Of Glutamic Acid Fermentation Process Based On Prior Knowledge And Improved LS-SVM

Glutamate ferment process highly non-linear and dynamicity, and its internal mechanism are very complicated. To realize the process of glutamate fermentation optimization and control, getting enough information of fermentation process is necessary. But the absence of reliable sensors for process variable on-line inspection, therefore, exploring appropriate intelligent modeling method, and applied the glutamate fermentation has become an important research direction.Researchers have long put forward manifold glutamate fermentation process of modeling methods. Such as mechanism modeling method, using neural network model method and support vector machine (SVM) method, etc. Mechanism model needs a large number of empirical knowledge accumulation and substantial computation, setting up relative trouble, and the accuracy is not high, Neural network theory based on empirical risk minimization principle, easy appeared fitting, local minima; Based on structural risk minimization principle Support Vector Machine (SVM) method, Support Vector Machine (SVM) overcame the neural network method such as inherent shortcoming, greatly improving the model generalization ability.First, the glutamate fermentation process mechanism is analyzed, and a classic glutamic acid fermented mechanism model is selected according to master knowledge. The models constructed and parameters identification is based on the initial experimental data. Four-Runge-Kutta method is used to calculate the corresponding moment of mycelium concentration, glutamate concentration and substrate concentration, and compare them with experimental data, analyzing model errors.Secondly, a state estimation model is set up by support vector for glutamate fermentation and the effect of model by SVM's parameters is studied. The support vector machine approach's advantages and disadvantages are pointed out by compared with the traditional radiate basis Function (RBF) neural network method.Thirdly, because of the short of the standard of the algorithm of Support Vector Machine, The Least squares Support Vector machines (LS-SVM) is used to establish glutamate fermentation state estimation model, and it is compared with the standard Support Vector Machine (SVM)model. Results show that the LS-SVM keep SVM modeling algorithm performance and the speed is due to standard SVM more conducive to online model to estimate.Finally, a hybrid modeling combined mechanism model method and support vector machine (SVM) method is putted forward based on mechanism the precision of the model is not high, and SVM and LS-SVM state prediction model is hard to adapt to the fermentation process different growth deficiencies. Mechanism model on the mechanism of reflection and support vector machine stronger generalization ability are contained in the mixture model. Experimental results indicate that the proposed hybrid support vector machine (SVM) model with excellent performance and improve the single SVM or LS-SVM modeling accuracy. Sensor module of glutamic acid fermentation process is constructed with MATLAB in VC + + programming hybrid soft based on several models.Aiming at a great impact on model effect of model parameters, and a genetic chaos optimization method putted forward, improved model accuracy.