A physics-guided integrated machine learning model for flow boiling pressure drop prediction in horizontal tubes

The flow boiling pressure drop is affected by the coupled effects of fluid properties and operating conditions. Traditional correlations and single machine learning (ML) models often showed limited generalizability across different conditions. An integrated ML model was developed based on physical mechanisms. To support model training and validation, this study collected experimental data from 32 published papers and established a database containing 2522 data points. Based on the criterion Bo * Re l 0.5 = 200 previously proposed for classifying conventional and micro/mini-channel, Categorical Boosting (CatBoost) regression models were established for the two channel types(CatBoost I for micro/mini-channel and CatBoost II for conventional channel). The Tabular Prior-Data Fitted Network (TabPFN) model was used to estimate the probability of a sample belonging to different types, and the final prediction was a probability-weighted fusion of the predictions from the two models. To assess the model’s internal generalization ability, this study used bootstrap-based method to estimate optimism-corrected performance on the training data. Comparisons against the baseline models and traditional correlations demonstrate that the proposed model exhibits strong predictive performance. The integrated model achieves a coefficient of determination (R 2 ) of 0.928 and a mean absolute relative deviation ( e a ) of 14.84% on the external dataset, outperforming other models. Since approximately 99.76% of the data corresponds to Re g > 2000, the applicability of the model is primarily validated within this range. In summary, these results demonstrate that the physics-guided integrated ML model proposed in this paper provides an effective method for predicting pressure drop.