Adaptive neural PID control case study:: Tubular chemical reactor

A new simple solution of deriving adaptive PID control based on artificial neural network (ANN) policy is presented. The design problem consists of training of both, a recurrent and a feed-forward neural network model. The recurrent and the feedforward neural network are respectively used as a predictor and adaptive neural feedback controller. They are trained using the back-propagation through the time and stochastic approximation algorithms, respectively. The multi-layer feedforward ANN is trained so as to achieve the control objective. The controller consists of a multi-layer feedforward ANN and a discrete-time PID algorithm. The network serves for adaptive tuning of the PID controller. To demonstrate the feasibility and the performance of this control scheme, a tubular chemical reactor is chosen as a realistic non-linear case study. Simulation results demonstrate the usefulness and the robustness of the control system proposed.