Comparison of static and dynamic neural network models in predicting outlet temperature of shell and tube heat exchanger plant

Abstract

This paper presents the comparison of two basic types of neural network (NN); static and dynamic, in predicting the outlet temperature of heat exchangers. Feedforward NN was used as static network while Time delay NN was used for dynamic network. Experimental data was collected from pilot-scale shell and tube heat exchanger in order to provide sufficient data processing i.e. training, testing and validation data to develop the models. The static and dynamic network models for the heat exchanger were developed in a Matlab® environment. The performances of the models were assessed through statistical validity by using the correlation coefficient and the means square error. For time series predictions, the dynamic neural network showed better results than the static neural network.