Abstract: Since the parameters in control system models are usually unknown in industrial applications, this paper tries to identify the system parameters by using the modified recursive prediction error algorithm for neural networks, and then design a self tuning PID controller via the pole-assignment method. Compared with the neural network identification based on the gradient learning algorithm and conventional PID, the method in this paper has simple structure of parameters, sustainable adjustment of neuron weights and quick calculation speed. Furthermore, this digital PID controller also enjoys good performance and easy application. And the simulation results verify that the effectiveness of this identification algorithm as well as the controller in this paper.

Key words: the modified recursive prediction error algorithm; neural network; self tuning PID via pole-assignment