Control of combustion processes based on optical variables can increase efficiency and reduce contaminant emissions in many industrial settings. Nowadays, optical variables such as total radiation Radt and flame temperature Tf are widely used for combustion diagnostics but not yet integrated into closed-loop control schemes, given by the combustion complexity and computational limitations. This paper advances toward a model-based control solution by finding a suitable structure for modeling the dynamic behavior of these optical variables. Evaluations using Akaike's Information Criterion and accuracy over field measurements show that Hammerstein and Wiener structures can accurately model both total radiation in a ladle furnace preheater and flame temperature in a coal boiler, considering SISO and TISO scenarios.