Pressure-dependent spectral-moment-constrained physics-informed neural-network WSGG model for multi-species gas mixtures

Radiative heat transfer in combustion over varying pressures is strongly affected by the nongray absorption of participating gases, making efficient and accurate radiation modeling essential for combustion simulations involving radiative heat transfer. This work develops a pressure-dependent, spectral-moment-constrained physics-informed neural-network weighted-sum-of-gray-gases (PINN-WSGG) model for H 2 O–CO 2 –CO mixtures. Total pressure is introduced as an explicit input, and Planck-weighted spectral moments are incorporated to constrain the gray-gas decomposition. The model covers temperatures from 300 to 3000 K, pressures from 0.1 to 80 bar, and arbitrary mole-fraction combinations of H 2 O, CO 2 , and CO. Validation against line-by-line-based one-dimensional benchmarks, conventional pressure-dependent WSGG models, and full-spectrum correlated- 𝑘 (FSCK) calculations shows that the model maintains high accuracy over this wide parameter range and improves radiative heat-loss predictions. Applications to a laboratory flame and an industrial gas turbine combustor further demonstrate reliable radiative heat transfer predictions with negligible spectral-parameter evaluation overhead. The proposed model provides a compact and practical approach for nongray radiation modeling under diverse combustion conditions.