This depends on how the DNI and DHI values are defined in your data.
In PVsyst, we have the relation (which is a definition):
GHI = DHI + BHI = DHI + DNI / Sin(Hsun)
DHI = Diffuse on horizontal plane,
BHI = Beam on horizontal plane,
DNI = Beam (or Direct) normal,
Hsun = sun's height, acc. to solar geometry).
Therefore for a given GHI, there is a bijective relationship between DHI and DNI: you can calculate one from the other, and there is no possibility of any discrepancy.
This is a question of definition, not a model.
In other words, the values (measurements) of DHI and DNI are equivalent.
NB: The determination of BHI from DNI/sin(Hsun) is very dependent on the Hsun value, especially for low sun heights (in the morning and the evening). That is, the time of the solar geometry is critical: the geometry should be calculated at the middle of the DNI measurement interval. If you observe discrepancies in the morning or the evening you have to carefully adjust the time shift parameter, already at the import stage.
Now if you have the same definitions as PVsyst and a correct time definition, discrepancies in your data are necessarily a measurement error.
For your simulations, you have to choose either the DNI or the DHI measurement, the one in which you have the best confidence.