The Modeling And Control Of The Fermentation Process Based On Fuzzy Neural Network

The fermentation industry is more and more important in national economy and the people's livelihood. It's operation and control are a very urgent problem. However, the fermentation process is distinctly nonlinear, its dynamics is not known precisely, and changes in initial conditions and variations of parameters with time, and the key state variables are not accessible. It is very difficult to apply traditional control methods to the complicated system. So, the modeling and control of the specific complex nonlinear dynamical systems have been studied systematically based on wavelet theory and nonlinear system control in this dissertation.1. The identification and control methods for the nonlinear systems is reviewed. Their features and using scopes are analyzed. The action and foreground of the intelligent control theory in the modeling and control of the nonlinear systems are discussed. Then, the internal and external current situation of study and existing problems in the modeling and control of the fermentation process are summed up.2. A hybrid approach for fuzzy system design based on fuzzy clustering and a kind of fuzzy neural networks. Firstly, Fuzzy cluster is used to divide fuzzy space and generate an initial fuzzy rules base from the given input data. Secondly, Adaptive-Network based Fuzzy Inference Systems(ANFIS) turn membership functions in antecedent part and parameters in consequent par of inference rules by means of least-square methods and gradient descent methods. Simulation results show that This fuzzy neural networks(FNN) has better generalization and approximation abilities.3. It is compared the three kinds of multi-resolution neural networks (MRNN). The training of the multi-resolution neural networks in parallel is simple. But the generalization ability is bad compared to the other multi-resolution neural networks. So we use the multi-resolution fuzzy neural networks model. It has the better generations ability andconvergence.4. It is used fuzzy neural networks (FNN) of the intelligent theory to control the formation process. In order to realize the real-time control, we use the compensatory learning algorithm. Simulation results show that it has the smaller and the sooner response rate, and it is robust to vary of parameters with time.Finally, all the results are summarized, and the study prospect is discussed.