Neural Network based Model Reference Control for Electric Heating Furnace with Input Saturation

Abstract

In this paper, a recurrent neural network based controller is designed for an electric heating furnace with input saturation. First, a recurrent neural network is used as a mathematical model of the electric heating furnace. The network is trained with training data of power in percentage and furnace temperature measured from the furnace. Then, this network is applied to design a recurrent neural network controller based on the model reference control method, which guarantees the input constraint of the plant. The plant network and the controller network are combined to form a closed neural network. During the closed neural network training, only parameters of the network controller are updated, and that of the plant network are kept unchanged. Experimental results show that the neural network controller provides the same performance as the PI and MPC controllers, and it satisfies the input constraint.