Neural Network Inverse Control Method And Its Application In Biological Fermentation Process

Biological fermentation technology is one of the important technologies for economic development and plays an important role in many fields such as agriculture and chemical. There have higher demands for automation technology in fermentation industry with the expanding of manufacturing scale. It is important for improving the quality and quantity of products to achieve high quality control in biological fermentation process. Biological fermentation process that involves the growth of organisms is complex. It is difficult to obtain high control performance by using the traditional control method. Neural network inverse (NNI) control method is presented which combines inverse system and neural network method. It is independent of precise mathematical model and applicable for fermentation process which is uncertainties.NNI control method has been achieved considerable success, but the control performance still needs to be improved in the complex industry process. The improved neural network inverse control methods are investigated to achieve better control performance. The main contents of this thesis:1. Online learning NNI control method is proposed. The initial values of online learning neural network are the parameters of neural network which is trained offline. According to the error between NNI system inputs and the original system outputs, a learning algorithm for neural network is designed based on the basis function theory, and then the convergence of neural network is analyzed. When the parameters of the original system vary, neural network does not need to be retrained again. The weights of neural network can be adjusted to reduce the inverse system inverse error and keep good control performance. The proposed control method satisfies the real time requirements of process control.2. Adaptive feedback compensating control method based on NNI system is proposed. A pseudo-linear composite system can be obstained by cascading neural network well trained. Considering the effect of neural network inversion error on the control performance, adaptive feedback compensating controller is designed to eliminate the inversion error. It can improve the stability of controlled system and decoupling performance of NNI system. The initial values of adaptive compensating controller are the connective weights of neural network which is trained well to estimate the inversion error. The parameters adaptation rule is derived from Lyapunov stability analysis and guarantees that the parameter estimation errors and the tracking errors are bounded.3. Model free adaptive control method based on NNI system is proposed for the multivariable coupling nonlinear system. A pseudo-linear composite system which includes multiple independent subsystems can be gotten by cascading NNI system and the original system. The uncertainties such as NNI modeling error and outside distraction can be regarded as the weak point of every subsystem. Then model free adaptive control method is designed for the subsystem. Input and output information are used in designing of model free adaptive control and the controller structure and parameters are adaptive. The proposed control method has the advantages of simple structure, good robustness and easy to implement.4. NNI control method and its three improved methods are applied in biological fermentation decoupling control problem. The numerical simulation results show that the improved NNI control method is effective, feasible, and has good control performance.