Predicting outdoor performance of solar parabolic trough collector (PTC) from indoor test data using machine learning

This study presents a proof-of-concept for using machine learning models to predict the outdoor thermal efficiency of Parabolic Trough Collectors based on standardized indoor heat loss test data. This approach aims to offer a cost-effective alternative to resource-intensive outdoor performance testing and address the difference between indoor and outdoor results. Using ISO 9806 standard, experimental indoor and outdoor data for a small-scale PTC receiver (UF-RT01) were analyzed. Predictive models (Least Squares Regression, Random Forest (RF), Extra Trees (ET), and both Simple and Complex Artificial Neural Networks (ANNs)) were trained on baseline data (Dataset 1) and two augmented datasets varying optical efficiency (Dataset 2) and thermal performance (Dataset 3). The performance of the machine learning model was evaluated using Coefficient of Determination (R 2 ) and Mean Squared Error on both internal test split and a separate, independent evaluation dataset. A difference between indoor and outdoor performance (exceeding 5 percentage points at ΔT = 100 °C) was observed, which demonstrates a high degree of predictive accuracy. Least Squares Regression baseline model achieved an R 2 of 0.999 and a Mean Squared Error (MSE) of 2.9394 × 10 −7 on the experimental dataset, though it failed to generalize. Models trained only on baseline data showed overfitting (R 2 < 0), while training with augmented datasets improved generalization. The ET model trained on the most diverse dataset (Dataset 3) performed best overall (R 2 = 0.75). The Complex ANN showed strong performance on Dataset 2 (R 2 = 0.82) but inconsistent results on Dataset 3. RF and Simple ANN models failed to generalize reliably. These findings showed the feasibility of using machine learning for predicting outdoor PTC efficiency. However, broader applicability requires richer datasets and integration of physics-based features.