SHAP-BASED INTERPRETABLE ENSEMBLE LEARNING FOR PREDICTING HEAT TRANSFER PERFORMANCE IN MICROCHANNEL HEAT EXCHANGERS WITH LOUVERED FINS

Louvered fins play a crucial role in microchannel heat exchangers. Previous studies on louvered fins have largely relied on traditional experiments and computational fluid dynamics (CFD) simulations. However, these methods tend to be time-consuming and resource-intensive. Therefore, there is a need to develop a fast and accurate prediction method to save experimental costs and reduce simulation resource consumption, facilitating subsequent radiator optimization and sensitivity analysis of parameters. The contribution of this paper lies in the proposal of an ensemble learning model that combines the strengths of generalized regression neural network (GRNN) in learning the <i>f</i> factor and support vector regression (SVR) model in learning the <i>j</i> factor, integrating these techniques with a gradient boosting decision tree (GBDT) meta-learner. The model significantly improves prediction accuracy compared to individual GRNN and SVR models, with an R<sup>2</sup> value exceeding 0.99. Additionally, SHapley Additive exPlanation (SHAP) value analysis is performed on the ensemble learning model to highlight the impact of each feature on the output parameters. The proposed method provides valuable insights into predicting the heat transfer performance of microchannel heat exchangers.