Artificial Neural Networks for Performance Prediction of CO2 Transcritical Heat Pump System

The rising global demand for energy-efficient heating systems has increased interest in [Formula: see text]-based transcritical heat pumps due to their unique thermodynamic behavior and environmental advantages. However, conventional models struggle with the nonlinear nature of transcritical cycles, and existing artificial neural network (ANN) approaches often fail to capture key interacting parameters under varying conditions. This study presents an advanced ANN-based model to accurately predict the performance of a [Formula: see text] transcritical heat pump, validated against experimental data. A feed-forward neural network is developed to analyze nonlinear system relationships using six input parameters: initial charge state, throttle valve opening, evaporator and gas cooler water flow rates, and their respective temperatures. Five performance metrics are predicted: coefficient of performance, cooling capacity, heating capacity, compressor work input, and discharge pressure. The model is trained and validated using experimental data from 36 operating conditions, with dataset size determined through statistical power analysis to prevent overfitting. High prediction accuracy is achieved, with [Formula: see text] values ranging from 0.956 to 0.986 and MSE between [Formula: see text] and [Formula: see text]. Compared to conventional methods, the ANN model is more accurate, efficient, and practical and can be extended to optimize and analyze various transcritical [Formula: see text] system configurations.