Comparative Study of Different Machine Learning Models for Heat Transfer Performance Prediction of Evaporators in Modular Refrigerated Display Cabinets

Abstract

This study explores the potential of machine learning models to predict evaporator heat transfer performance in Modular Refrigerated Display Cases (MRDCs). Six experimental datasets from MRDC systems were analyzed to compare the efficacy of six machine learning models: Linear Regression, Decision Tree Regression, Support Vector Machines (SVMs), Feedforward Neural Networks (FNNs), Random Forest (RF), and Light Gradient Boosting Machine (LightGBM). The findings indicate that the ensemble tree-based models, LightGBM and RF, are particularly effective in predicting evaporator heat transfer performance. These models demonstrate high accuracy and robustness, effectively capturing the nonlinear relationship between the evaporator temperature and heat transfer coefficient. Moreover, LightGBM and RF exhibit notable stability and adaptability in scenarios of limited data availability and elevated noise levels. Their consistent predictive accuracy across different experimental conditions highlights their suitability for complex refrigeration systems. This research provides essential insights for optimizing MRDC evaporator performance, establishing a theoretical and data-driven foundation for energy-efficient enhancements and intelligent management within cold chain systems.