The Variable Refrigerant Flow (VRF) system uses mechanical components and control technologies, including variable speed compressors and electronic expansion valves, to enable precise control of refrigerant flow into each indoor unit based on the ther...
The Variable Refrigerant Flow (VRF) system uses mechanical components and control technologies, including variable speed compressors and electronic expansion valves, to enable precise control of refrigerant flow into each indoor unit based on the thermal load. Therefore, the VRF system allows individual control of indoor units and flexible operation, resulting in enhanced energy efficiency compared to traditional systems. With such advantages, the VRF system becomes one of popular HVAC systems. As VRF systems are widely used, accurate modeling and performance prediction have become crucial for efficient operation. Developing a model that accurately describes the operational characteristics of a VRF system has been challenging due to their complex operating mechanisms and manufacturer-specific control technologies. Although manufacturers provide product specifications and performance curves to some extent for reference, they often only disclose limited data that meet regulatory measurement standards while keeping detailed information about system control algorithms confidential to protect their proprietary technology.
Previous research approaches have attempted to address those limitations by utilizing either empirical regression models based on manufacturer performance data or physics-based models using building energy analysis tools such as EnergyPlus and TRNSYS. However, these modeling approaches are subject to limitations, including uncertainty issues and modeling complexities. Additionally, the performance of those models heavily relies on rated operational data, thereby necessitating the use of additional techniques to accurately reflect the actual operation of VRF systems. Meanwhile, conventional data-driven approaches face challenges when historical data is insufficient or unavailable, representing the need for the development of data-driven models and techniques to compensate for the limited information.
In this study, a method is proposed to develop a performance prediction model that improves upon previous modeling efforts by utilizing manufacturer's performance data and a small amount of field measurement data. The approach of this study focuses on accurate prediction of system performance while considering the actual operation of VRF systems. The proposed modeling method utilizes corrected manufacturer's rated performance data, which encompasses representative system information, as the foundation for the proposed model.
By preprocessing the rated performance data, we develop a power consumption prediction model that demonstrates reasonable performance under steady-state conditions. However, this model has limitations when applied to unsteady-state conditions. To address this, we present a method that incorporates actual operating conditions using a small amount of measurement data and calibrates the model performance based on the rated performance data. Additionally, the proposed method includes estimating cooling capacity by analyzing the relationship between total capacity (TC) and other variables in the rated performance data. This approach is necessary as obtaining cooling capacity data through field experiments poses significant challenges.
For model development, the XGBoost algorithm is employed that is one of a gradient boosting decision tree algorithm, and the model hyperparameters are opimized using random search cross-validation. Finally, by conducting case studies, the performance and applicability of our proposed modeling method are demonstrated.
In summary, this study contributes to the advancement of modeling methods in VRF system performance prediction and supports efficient operation.