Just to add that most empirical methods (nomograms or simple parametric equations) are not sufficiently general to accurately determine forging load for a wide range of parts (and materials).
If the empirical result is to be compared. I would say 20% is good enough for the first comparison between the simulation result and the result from experiment. This means the simulation result might be 20% more or less than the empirical result. However, what I would do then is to adjust and/or validate the correctness of related parameters such as the friction coefficients, boundary conditions, thermal conditions, flow stress and number of meshes, etc. The difference should be reduced after some reasonable adjustments.