Identification of the relevant input variables for predicting the parabolic trough solar collector's outlet temperature using an artificial neural network and a multiple linear regression model

Abstract

The main objective of this study is to present the most influencing input variables for a parabolic trough solar collector (PTSC) outlet temperature through prediction and optimization. Six artificial neural network (ANN) and four multiple linear regression (MLR) models were proposed, validated, and compared in detail. Temperature, wind speed, rim angle, flow rate, and solar radiation were used as input variables. The simulation showed that ANN-1 and MLR with Second-Order Equation (SOE) are the models that yielded the best results with R2 = 0.9984 and R2 = 0.9958 and with an RMSE = 0.7708 and 1.6031, respectively. The sensitivity analysis results of the ANN-1 model trained, with and without biases, showed that the inlet temperature was the most significant parameter influencing the PTSC outlet temperature. Both models yielding the best results were inverted to estimate the optimal input parameter using the trust-region reflective algorithm optimization method. The optimization results showed that ANNi and MLR-SOEi estimated the input temperature with an error < 4.008% and had a very short-elapsed prediction time <0.2277 s. Due to high accuracy and short computing time, ANN-1 and ANNi are more suitable than MLR-SOE for simulating and optimizing the PTSC outlet temperature. Likewise, the MLR-SOE method proved to be a simpler and cheaper alternative than the ANN method.