Design of neural network PID controller based on E-FRIT

Abstract

Proportional–integral–derivative (PID) controllers have been widely used for process systems. However, a good control result is not always obtained with fixed PID gains when a controlled object has nonlinearity. This paper proposes a design method for a nonlinear PID controller that utilizes a neural network to overcome the problem. In the proposed controller, PID gains are tuned online by a neural network and a controlled object is manipulated by the PID controller with the tuned PID gains. The neural network is learned by an offline learning algorithm based on the Extended Fictitious Reference Iterative Tuning (E-FRIT) and the backpropagation. E-FRIT is a method that tunes control parameters directly by using operating data and evaluates not only a controlled output but also the difference of manipulated variable. Simulation examples are provided to show the effectiveness of the proposed method. Moreover, the experimental result of a level control of a tank system is also given to demonstrate the performance of the proposed method.