Comparison of Alternative Approaches to Neural Network PID Autotuning

摘要

In this paper, a scheme for the automatic tuning of PID controllers on-line, with the assistance of trained neural networks, is proposed. Two alternative approaches are presented and compared. The two approaches differ mainly on the plant identification measures employed. In the first one, applicable in open and closed-loop cases, the plant is identified using integral measures of the output or the control signal. The second approach, applicable only in closed loop, identifies the plant by its critical frequency and gain. In both cases, the neural networks approximate the mappings between the plant identification measures and the corresponding optimal PID parameters, in the ITAE sense. Simulation results show that, in noise free situations, similar tunings are obtained by both approaches, much better damped than those obtained by the application of Ziegler-Nichols tuning rule. Preliminary results show that the second approach offers a more robust performance in the presence of noise.