Prediction Of The Outlet Temperature Of An Experimental Heat Exchanger Using Artificial Neural Networks
Free (open access)
213 - 220
A. Tlatelpa-Becerro, L. Castro-Gómez, G. Urquiza, R. Rico-Martínez
Artificial neural networks (ANNs) coupled with a Kalman filter were applied to predict the outlet temperatures of a laboratory crossflow heat exchanger. Time series of inlet temperatures were obtained and used as input values for training and testing the ANN. The ANN showed good prediction capabilities, but the high noise presented in the inlet variables frequently prevented the ANN prediction to match the evolution observed in the experiment. In order to further improve the tracking capabilities of the ANN, a Kalman filter was introduced. The results of the assisted scheme showed the maximum deviation between the predicted results and experimental data with relative errors below 0.007 y 0.002% for the water and air outlet temperature respectively. This assisted model reference strategy can be applied for thermal analysis in engineering applications and the development of predictive control applications in real time.
artificial neural networks, time series, Kalman filter, heat exchanger