Artificial neural network-based psychrometric predictor

An artificial neural network (ANN)-based psychrometric chart could be used for real-time calculations of the air properties required in drying of agricultural and food materials, and ventilation of farm buildings. Two ANN were developed to predict psychrometric parameters. In the first ANN, dry-bulb temperature t(dp) and relative humidity rho were inputs, and dew point temperature t(db), wet-bulb temperature t(wb), enthalpy h, humidity ratio W. and specific volume v were outputs. In the second ANN, t(db) and t(dp) were inputs and t(wb), rho, h. W and v were outputs. The data used to train and verify the ANN were obtained from psychrometric mathematical models. Reasonable accuracy was obtained for all predictions for practical applications. Shrinking the range of predicted variables using mathematical functions improved the ANN accuracy. In the, predictions with relative errors <5% for t(dp), t(wb), h, W and v were >93.0, 95.8, 95.4, 95.9 and first ANN. 99.9% of total, respectively. In the second ANN, predictions with relative errors <5% for 9, t(wb), h, Wand v were >95.7, 97.0, 92.4, 99.5 and 100.0% of total, respectively. (C) 2003 Silsoe Research Institute. All rights reserved Published by Elsevier Science Ltd.