Time-Delay Temperature Control System Design based on Recurrent Neural Network

A Recurrent Neural Network (RNN) is a special neural network sequence model that is very suitable for dealing with time series tasks. In various industrial processing systems, it has achieved good performances. In this paper, a RNN model which is driven by an ideal reference model is proposed for the single-input single-output(SISO) temperature control system with time-delay. An ideal reference model is introduced to provides a more valuable teaching signal for helping RNN controller to obtain higher learning efficiency and providing suitable control input to the temperature control system. Meanwhile, Adam optimization algorithm which can get adaptive learning rates is used to update parameters and improve the control performance of the RNN. Further, a classical integral proportional derivative (I-PD) controller is designed to reduce the effects caused by the temperature setting value kick during the RNN learning period. Simulations were developed under the MATLAB environment to evaluate the proposed control system performance. In order to demonstrate the efficiency and application of the proposed RNN control method, the simulation results based on the actual temperature model are compared quantitatively.