Akademik Veri Yönetim Sistemi
Journal of Thermal Analysis and Calorimetry, cilt.150, sa.24, ss.20291-20301, 2025 (SCI-Expanded, Scopus)
In this study, the performance of an internal heat exchanger cooling cycle using environmentally friendly R454A and R454C refrigerants has been predicted using advanced machine learning (ML) methods. Although they have basic features, conventional thermodynamic models often require iterative calculations and various assumptions. This situation makes it difficult to accurately reflect the nonlinear relationships between variables such as evaporator and condenser temperatures, refrigerant properties, and IHX efficiency. To overcome these limitations, this study combines six different ML algorithms: Support Vector Regression (SVR), Extreme Gradient Boosting Regressor (XGBR), K-Nearest Neighbors (KNN), Random Forest, Light Gradient Boosting Machine (LGBM), and Categorical Boosting (CatBoost). These methods are used to predict critical energy and exergy indicators such as the Coefficient of Performance (COP) and Second Law Efficiency (ηII). Both the XGBR and CatBoost models are notable for their exceptional prediction performance, with R2 values reaching 0.9993 for energy analysis and 0.9986 for exergy metrics, demonstrating that these models can effectively capture the nonlinear relationships between thermodynamic variables (e.g., evaporator and condenser temperatures) and key performance indicators such as COP and ηII. In addition, CatBoost achieves < 1% MAPE levels, offering accuracy comparable to XGBR in applications where MAPE minimization is critical, thus providing engineers with a flexible option that does not compromise on prediction power. The study’s findings will provide a solid foundation for future research on the use of ML in refrigerant cycle optimization. The employment of ML will enable engineers to validate model logic with domain knowledge and improve system designs.