Physics-Informed Machine Learning–Based Energy and Exergy Performance Prediction of a Vapour Compression Refrigeration System Operated with a Low-GWP Refrigerant

Precise modelling of vapour compression refrigeration system (VCRS) performance is crucial for enhancing energy efficiency and for the adaptation to low global warming potential (GWP) refrigerants. Conventional thermodynamic models make more broad assumptions; while purely ML-based machine learning approaches are typically less physically consistent and interpretable. In this work, this study proposes a physics-informed machine learning framework combining energy and exergy analysis with supervised regression techniques for model performance, compressor power consumption, and exergy efficiency prediction in a VCRS model under different operating conditions. Experimental studies were conducted by varying compressor speed, evaporator temperature, and condenser temperature. A baseline thermodynamic model employing first- and second-law principles was constructed and machine learning models are trained to predict residual discrepancies between measured values and thermodynamic predictions. We have evaluated Gradient Boosting Regression, Random Forest Regression, and Support Vector Regression. The hybrid physics-informed framework achieved coefficients of determination of over 0.97 for both the coefficient of performance and exergy efficiency which compared better with standalone thermodynamic models and the pure data-driven methods. The integrated exergy analysis helped to reveal predominant drivers of irreversible processes, and the physics-informed framework made it possible to predict physically plausible behaviour over the range of operating conditions.