Machine learning approach for the prediction and optimization of heat transfer performance of ZnO/TiO₂-R123 in ORC evaporator

This study employs machine learning approaches to develop a back propagation (BP) model for capturing flow boiling heat transfer characteristics of ZnO/TiO₂-R123 in a horizontal tube and conducts a bi-objective optimization considering heat transfer performance and flow resistance simultaneously. The BP neural network model is established with 750 groups of experimental data as training samples and 150 groups of experimental data as testing samples. The training accuracy and predictive accuracy of the BP model are analysed in detail. The effects of six operation parameters on heat transfer coefficient and pressure drop are examined, along with a bi-objective optimization conducted to maximize heat transfer coefficient and minimize pressure drop. The results indicate that the BP neural network model achieves a very high prediction accuracy, with a relative error of ±1.5 % for the prediction of heat transfer coefficient pressure drop. The heat transfer coefficient is negatively correlated with the vapor quality and increases slowly with the mass flux. The pressure drop increases slightly with the outlet temperature and decreases slowly with the outlet pressure at first, then gradually becomes steeper. The optimal solution for heat transfer coefficient and pressure drop are 4500 W/(m²·K) and 0.022 MPa, respectively.