System Identification of a Cooling Coil Using Recurrent Neural Networks

Modelling a non-linear plant by employing first principles is a tedious task and the resulting model often does not replicate the true behaviour of the plant. The cooling coil of an air-handling unit exhibits severe non-linearity and it is difficult to develop a first-principle model which can replicate the true plant behaviour. The paper discusses the modelling of a cooling coil of an air-handling unit using first-principle approach as well as artificial neural networks. In order to capture the dynamics of the cooling coil, a type of recurrent neural network, i.e. non-linear autoregressive network with exogenous input (NARX) is used. It has been demonstrated that a recurrent neural network with sufficient training has better performance as compared to the model obtained using first principle.