Research On Control Model And Algorithm In Dynamic Systems Based On Neural Networks

The field of automatic control has consummately solved the theory and method of the linear system and gained some achievements for the complicated nonlinear system. However, for the complexity and time-varying of actual system, especially for some uncertainty of the process, it is difficult to establish the precise mathematical model for description. Therefore, the traditional methods have the universal problems of difficultly model, the low accuracy and difficultly solve so on. In recent years, the theory and method of intelligent control have been attached importance and developed, which provides the effective way for the control problem of complex nonlinear dynamic system. Because neural networks have highly nonlinear mapping capability, large-scale parallel distributed processing and good adaptive learning mechanism. There is great potential in the identification and control of the dynamic system for it. Therefore the study for application of artificial neural network to the identification and control has good adaptability.The paper analyzes the current common method used in identification and control. For the problems of identify control and optimization to solve in dynamic system, it studies the neural network model and learning algorithm. It combines the neural network,fuzzy logic and evolutionary algorithm. The paper constructs different neural network model to achieve nonlinear system identification control and process optimization to solve. In the system identification, it uses the most typical BP neural network to the studies of nonlinear system identification. It uses the fuzzy neural network to the design of control model. It takes the improved quantum genetic algorithm as optimal solution for the parameter of the control model.Traditional neural network builds based on traditional neurons, it is a static model in its essence. It is difficult to reflect the time cumulate effect of process input about dynamic system. The paper takes the process neural network as the identification and control of the dynamic system and presents the control model based on feed forward process neural network, builds the forward model and backward model. The experiment proves that this model has good practicality.Elman network has been widely used in the aspect of the system identification and control. For the dynamic system that input and output are time-varying functions, it has its own limitations. The paper presents the Elman feedback process neural network which input and output are time-varying functions. In comparison with the feedforward neural network, it has the characteristics of the high study efficient, fast rate of closure and can realize the system identification and control of nonlinear system of higher order.The paper presents an improved quantum genetic algorithm and applied to solve the parameter optimization for feedforward process neural network and fuzzy neural network. In comparison with the quantum genetic algorithm and the algorithm of gradient descent based on function basis expansion, this algorithm possesses the stronger advantage. Further, in the training of process neural networks, the conventional Newton method has heavy computation. Improving by BFGS Quasi-Newton, the training speed has been greatly improved.