Deep learning based rapid calculation approach for gas radiation characteristics considering foreign gas collision broadening

Abstract

The line-by-line method(LBL) is the most accurate radiation calculation method, but it is difficult to practically apply due to its complicated and time-consuming calculation process. The heterogeneous particle collision broadening of foreign gas must be considered under mixed gas conditions. Simply replacing with air broadening will bring unexpected errors. A rapid radiation calculation method based on back-propagation neural network (BPNN) is proposed in this paper. This method fully considers the influence of foreign gas under mixed gas conditions. It has the advantages of high calculation efficiency, accuracy equivalent to the line-by-line method, and small storage space required. In this paper, the calculation of the radiation characteristics of a mixture of CO2, H2O, and air near 2.7 μm is studied as an example. Three methods of LBL, BPNN, and LBL which use air broadening instead of specific gas broadening (LBL(A)) are compared. The results show that LBL(A) has an obvious error under some conditions. Meanwhile, BPNN can be 3 orders of magnitude faster than LBL with relative error less than 0.2%, and the storage space is only about 6 M.