Intelligent prediction of solar air heater performance using hybrid XGBoost-multi-layer perceptron machine learning framework

This investigation presents a innovative hybrid machine learning framework combining Extreme Gradient Boosting (XGBoost) and Multi-Layer Perceptron (MLP) neural networks for accurately predicting the thermo-hydraulic performance of roughened solar air heaters equipped with twisted rib roughness. This study addresses the critical research gap in systematically comparing advanced machine learning approaches for performance prediction. The meticulously designed experimental setup investigated key parameters: P/e = 6–10, α = 30°-90°, y/e = 3–7 and Re = 3,500–21,000. The hybrid framework strategically leverages XGBoost’s exceptional structured feature extraction capability and MLP’s remarkable proficiency in capturing complex non-linear relationships, with intelligent feature scaling ensuring optimal learning convergence. The model demonstrated exceptional predictive accuracy as Nu (R 2 = 0.992, MAE = 1.78), Fr (R 2 = 0.857, MAE = 0.169), and THP (R 2 = 0.802, MAE = 0.0031). Remarkably, 94.3% of predictions fell within ±5% of experimental values. Optimal configurations were identified at P/e = 8, α = 60°, and y/e = 3 with Re = 21,000, achieving maximum THP of 2.136. This synergistic hybrid approach successfully overcomes conventional single-model limitations, providing a computationally efficient, highly accurate alternative to time-consuming experimental investigations, thereby significantly contributing to sustainable energy system optimization.