Improving the accuracy of predicting the performance of solar collectors through clustering analysis with artificial neural network models

Abstract

Accurate prediction of collector performance is important for optimal planning of solar thermal systems. Here, a novel prediction method combining clustering of data with artificial neural network (ANN) model is presented. A novel all-glass straight-through tube solar collector is employed as reference solar technology. In the present approach, experimental collector performance data was first collected during different weather conditions (sunny, cloudy, rainy days) subject to a clustering analysis to screen out outlier samples. The data was then used to train and verify the neural network models. For the ANN, the Back Propagation (BP) and Convolutional Neural Network (CNN) models were used. For predicting the performance (thermal efficiency) of the solar collector, solar radiation intensity, ambient temperature, wind speed, fluid flow rate, and fluid inlet temperature were used as the input parameters in the model. The prediction accuracy of the neural network models after full-data-screening were superior to that of the pre-screening and partial-screening models. The CNN model provided somewhat better efficiency predictions than the BP model. The R2, RMSE and MAE of the CNN model prediction in sunny conditions with full-screening was 0.9693, 0.0039 and 0.0030, respectively. The average MAPE of the BP and CNN models for all three weather types dropped by 30.7% and 13.8%, respectively, when applying data pre-screening and partial-screening only. The accuracy of the ANN collector prediction model can thus be improved through data clustering, which provides an effective method for performance prediction of solar collectors.