Hot water temperature prediction using a dynamic neural network for absorption chiller application in Indonesia

Weather condition particularly for solar radiation and dry bulb temperature has important role in absorption chiller performance. In this paper hot water temperature prediction in generator inlet of absorption chiller has been conducted under various weather conditions. Dry bulb temperature and global horizontal radiation are selected as predictors. Three artificial neural network (ANN) types including feed forward back-propagation, cascade forward back-propagation, and Elman back propagation models have been investigated for prediction. Moreover, numbers of neuron and time delay effects were analyzed to achieve an accurate prediction. The results show that hot water temperature in generator inlet can be predicted precisely using a feed forward back propagation neural network with the configuration of a three hour delayed input on radiation, current dry bulb temperature, seven neurons, tan-sigmoid transfer function and Bayesian regularization algorithm. The prediction results perform a good agreement between predicted and experimental values. The error resulting from training and validation is 3.1 degrees C and 2.6 degrees C with a coefficient of variation at 4.4% and 3.5% respectively.