Analysis of transfer functions and normalizations in an ANN model that predicts the transport of energy in a parabolic trough solar collector
E.D. Reyes-Téllez, A. Parrales, G.E. Ramírez-Ramos, J.A. Hernández, G. Urquiza, Miguel Heredia, Fernando Sierra
Abstract
ABSTRACT Artificial neural network model was developed to obtain the fluid outlet temperature of the parabolic trough collector (PTC) with a grooved absorber tube. To improve the accuracy of the model, an analysis of the transfer functions was performed at different normalization intervals to choose the best model capable of estimating the PTC outlet temperature. A first model was developed that was used as a base model from 1,155 concordant experimental data. The variables were input temperature of the fluid, ambient temperature, irradiance, hour, day, configuration, volumetric flow, feeding temperature, and storage temperature. To validate the accuracy and the adaptability of the model proposed, statistical tests (coefficient of determination (R 2 ), root mean square error (RMSE), and mean absolute percent error (MAPE), significance test (F-Fisher and t-student), and linearity test (slope-intercept)) were performed. The base model was validated, having an R 2 = 0.9961, RMSE = 0.14706, MAPE = 0.00795, also approved significance and linearity tests. Consequently, ten models were developed for the analysis of the proposed normalization intervals using the same architecture as the base model at five areas of interest with a hyperbolic tangent sigmoid (TANSIG) and log-sigmoid (LOGISIG) functions. The results show that a model has higher accuracy; this model was the TANSIG of [0.1,0.9] with R 2 = 0.9974, RMSE = 0.12123, MAPE = 5.93 × 10 –5 , and approved significance and linearity.