Low global warming potential R1234yf in a mobile air-conditioning system: a study on performance prediction using different machine learning approaches
Machine learning (ML) approaches have admirable potential to forecast the performance of the mobile air-conditioning (MAC) system with low global warming potential R1234yf instead of conventional mathematical and simulation approaches. In this work, three different ML algorithms—artificial neural ne...
Gespeichert in:
Veröffentlicht in: | Journal of thermal analysis and calorimetry 2024-12, Vol.149 (23), p.14415-14432 |
---|---|
Hauptverfasser: | , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | Machine learning (ML) approaches have admirable potential to forecast the performance of the mobile air-conditioning (MAC) system with low global warming potential R1234yf instead of conventional mathematical and simulation approaches. In this work, three different ML algorithms—artificial neural network (ANN), simple recurrent neural network (SRNN), and extreme gradient boosting (XGB)—have been employed for predicting the energy and exergy performance. Compressor speed, condenser-side air velocity/temperature, and evaporator-side air flow rate/temperature were considered as influencing input parameters. In energy analysis, performance indexes, namely refrigerant flow rate, cooling capacity, compressor power, and coefficient of performance (COP), were considered as output parameters, while total exergy destruction and exergy efficiency (η
ex
) were accounted for as exergy metrics. First, the heat mapping method was used to rank the correlation among the input and output factors, and results revealed that compressor speed and evaporator-side air temperature are identified as the most and least influencing parameters on the forecast of energy and exergy performance metrics. Among the three models, the use of the XGB model showed excellent prediction efficiency on COP and η
ex
with root-mean-squared error of 0.0756 and 0.9786, respectively, while the corresponding correlation coefficients were 0.9749 and 0.9119. Predicting η
ex
using ANN and SRNN showed weak performance with a determination coefficient less than 0.70; moreover, prediction performance on energy indexes using ANN and SRNN models was good and almost identical. Overall, it is inferred that using XGB over ANN and SRNN can deliver superior prediction efficiency with enhanced reliability and can be employed as a forecasting platform for MACs under widespread working conditions. |
---|---|
ISSN: | 1388-6150 1588-2926 |
DOI: | 10.1007/s10973-024-13715-2 |