ANN controller trained with steady state input-output data for a heat exchanger

Abstract

This paper discusses the design and implementation of an Artificial Neural Network (ANN) based adaptive controller for a heat exchanger. The control strategy chosen is that of explicit nonlinear model predictive control. The nonlinear inverse model of the plant is identified from steady state input-output data by off-line training of a multilayered neural network through error back propagation. For performance enhancement, manipulation of training data and on-line parameter updating are tried. Single pass of derivative of error measure across the plant, on-line gave an excellent performance for regulatory as well as servo problem. The proposed cont roller is found to be successful over a wide operating range. The results are compared with that of an optimized PID controller.