Research On Adaptive Control Method Based On Neural Network System And Parameters Identification

All the time, how to ensure the controlled system runs steadily、controls accurately and adjusts sensitively, is a multiple problem in the automatic control field. At present, the convention PID control algorithm has been widely used to handle with the controlled system and has reached many great results, However, it has been proved that its role is very limited, which mainly shows in the following aspects. First of all, the convention PIDcontrol is not convenient to process mass abundance data by a modern digital computer; Second, for a class of dynamic, complex, time.varying, uncertain and nonlinear system, if only the PID control method is used, then it is very difficult to achieve the ideal control performance that we expected, such as high efficiency, high speed, super smooth, etc. To overcome the various technical defects that exist above, in this paper, the neural network control method and the pole assignment control method are used to design controllers for a class of nonlinear systems, namely as PID controller based on RBF neural network and PID controller based on pole assignment.In theory, the design of the above PID controllers depend on specific mathematical models can get some good results. But in fact, there exist many inherent requirements and exterior factors in practical engineering, like real.time adjustments and ever.changing working environments. Therefore, the model parameters or the controller parameters may change from time to time and so the online identification of system parameters must be prior to the design of controllers as well. Actually, various parameters (the model parameters and the controller parameters) in different algorithms are very difficult to be set. For example, even though the traditional methods like least squares estimation algorithm and kalman filter identification algorithm can be used in time.varying system, but once they are put into use in time.varying or the uncertainty and high order nonlinear system, then the self.adaptability of a controlled system would be reduced. Also, the adaptive PID controller may mot be able to provide the system with an accurate response, worse still, and the system may traps into a new round of adjustment, even instability or collapse directly.After considering the problems lie in the above adaptive PID controllers, analyzing the principle of the neural network structure and combining with the literature, this paper puts forward a new method of on.line parameter identification. The method is mainly based on predistortion processing. Or, more specifically, this identification method is:making the output signal of the traditional neural network structure predistortion processing and treating this processed output signal as an actual output, then comparing it with the desired output to get an error value, and the last, feeding the error back to the neural network structure. Broadly speaking, aiming at a real.time online parameters estimate, this article brings the improved recursive prediction error algorithm and the DTNN algorithm together, and then makes use of the gradient descent method to trim the network weights.Once the model parameters are got, in nature, a mathematical model, which is used to design the controller, is formed.In addition, in order to realize this project on the MATLAB simulation computation, this article also disposal of the transfer function discretization on.line. So after that, a new discrete mathematical model establishes though parameter identification. Finally the simulation experiment shows the effectiveness and validity of this method.