Adaptive neurocontrol of heat exchangers

This paper investigates the use of adaptive artificial neural networks (ANNs) to control the exit air temperature of a compact heat exchanger. The controllers, based on an internal model control scheme, can be adapted on-line on the basis of different performance criteria. By numerical simulation a methodology by which the weights and biases of the neural network are modified according to these criteria was developed. An ANN controller for an air-water compact heat exchanger in an experimental facility is then implemented. The parameters of the neural net are modified using three criteria: minimization of target error, stabilization of the closed-loop performance of the controller, and minimization of a performance index that we have taken to be the energy consumption. It is shown that the neural network is able to control the air exit temperature in the heat exchanger. The neurocontroller is able to adapt to major structural changes in the system as well as to simultaneously minimize the amount of energy used.