A Modified Approximate Internal Model-based Neural Control for the Typical Industrial Processes

Abstract

A modification of the Approximate Internal Modelbased Neural Control (AIMNC), using Multi Layer Neural Network (MLNN) is introduced. A necessary condition that the system provides zero steady-state error in case of the constant reference and constant disturbances is derived. In the considered control strategy only one neural network (NN), which is the neural model of the plant, should be trained off-line. An inverse neural controller can be directly obtained from the neural model without need for a further training. Simulations demonstrate performance improvement of the modified AIMNC strategy. An extension of the modified AIMNC controller for the typical industrial processes is proposed.