Building a PID controller using neural networks

R. A. Zhilov

News of the Kabardin-Balkar scientific center of RAS, no. 5 (2022), pp. 38-47

Résumé

The paper considers the use of neural networks to tune the PID controller. The need to use machine learning methods for tuning regulators stems from the complexity and duration of such tuning by a human. For each control object, a specialist has to adjust the PID controller coefficients, and in dynamic systems, they also have to be reconfigured. Also, the work assumes the use of hybrid neurocontrol systems and hybrid neural networks to simulate the operation of the PID controller itself. Recurrent neural networks are a powerful class of models that are well suited for modeling non-linear systems. One of the main applications of such neural networks is the control system. A sufficiently well trained recurrent neural network can simulate the operation of a PID controller. The advantage of this kind of controller is more accurate learning in conditions of only a fairly complete training set and the need for further adjustment by an expert. Also, replacing the PID controller system and the neuromodule with a hybrid neural network that performs the full work of this system simplifies it.

Détail

Keywords: hybrid neural networks, PID controller, neurocontrol, recurrent neural networks.