Nonlinear identification and control of a heat exchanger: A neural network approach

In this paper the potentials of neural networks-based control techniques are explored by applying a nonlinear generalized minimum variance control methodology to a simulated application example. In particular, reference is made to the control problem of regulating the output temperature of a liquid-satured steam heat exchanger by acting on the liquid flow-rate. Due to the non-minimum phase characteristic of the dynamics of the process, a simple inverting minimum variance controller is unsuitable. On the other hand, an effective solution is provided by a detuned model reference approach, which introduces a penalization factor in the control variable. A steady-state off-set error problem, caused by the neural network approximations, is tackled by means of an hybrid control structure, which combines a nonlinear integral action block with a neural controller. A comparison analysis is made to show the effectiveness of the proposed neural control schemes with respect to classical linear controllers.