The Study Of Controller Based On Neural Network

The general control methods can't meet the production requests in the industrial process because of the complex character and high request on the industry. So the thesis discusses using ANN control methods to meet the control request and improve control effect.The identification of the systems based on neural network is discussed. NN PID control and the composite control based on BP NN and GA are proposed based on the identification. The simulation experiments are made.The main works of this dissertation are summarized as follows:1.Because BP neural network's convergence speed is slow and it is easy to be trapped in local minima, the improved algorithm is given. The comparison between there is made.2.To the identification of the controlled object of the system, two kinds of identification methods are put forward, including BP neural network identification and RBF neural network identification. The methods to improve the generalization ability of neural networks are summarized. The deep discuss on the identification is made.3.According to the identification methods based on neural network, the control strategies is proposed. Another is the composite control based on BP NN and GA, which use the ability of the global optimization of GA to optimize the weights of neural network. The simulation results indicate this composite control can be in progress at overall and overcome the disadvantage that BP algorithm is sensitive to the initial weights value. This control can get better control results and improve the instability from the uncertainties, nonlinearity and large inertia in some industrial systems.The simulation results indicate that this composite neural networks controller , which has the STC structure based on the improved algorithm, can reduce overshoot and make system effective in robustness and anti jamming ,meanwhile prove the superiority of this controller to the PID controller.