A machine learning based efficient and compact full-spectrum correlated k-distribution model. (October 2020)

• Record Type: Journal Article
• Authors: Zhou, Ya; Wang, Chaojun; Ren, Tao
• Is Part Of: Journal of quantitative spectroscopy & radiative transfer. Volume 254(2020)
• Publication Date: 2020-10
• Subjects: Radiative heat transfer -- Machine learning -- Neural network -- Full-spectrum correlated-k-distribution -- Efficient and compact; Spectrum analysis -- Periodicals; Radiation -- Periodicals; Analyse spectrale -- Périodiques; Rayonnement -- Périodiques; Radiation; Spectrum analysis; Periodicals; 543.0858
• DOI: 10.1016/j.jqsrt.2020.107199
• Languages: English
• ISSNs: 0022-4073

Abstract

Highlights: A machine learning based efficient and compact FSCK model was developed. The machine learning model was trained with data from the previously developed FSCK look-up table. The model is only about 35 MB but is as accurate and efficient as the FSCK look-up table. Abstract: The full-spectrum correlated- k -distribution (FSCK) look-up table previously developed by Wang et al. (2018) provides an efficient means for accurate calculations of radiative properties in nonhomogeneous media, which was constructed for gas mixtures of three species, i.e., CO2, CO and H2 O within a large range of thermodynamic states. However, the total size of the FSCK look-up table is about 3.2 GB. In order to obtain accurate k -values for arbitrary thermodynamic states from the tabulated values, a 6-D linear interpolation method has to be employed. In practical applications, transferring and loading such a huge database are inconvenient and time-consuming. In this study, we presented a machine learning approach to accurately fit the FSCK look-up table, in order to provide an efficient and compact FSCK model. Without interpolation on the fly, correct k -distributions for any arbitrary thermodynamic states can be obtained directly and the total size of the trained machine learning model is only about 35 MB.