When you know the mathematical model of the disturbance. Therefore, you can compensate it in advance, instead of waiting until it affects the output and use the feedback loop to compensate.
When you know the mathematical model of the disturbance. Therefore, you can compensate it in advance, instead of waiting until it affects the output and use the feedback loop to compensate.
If you have a mathematical (nonlinear) model of your plant and, furthermore, the structure of this model allows a dynamic inversion, then I suggest to use a feedforward control. This will improve the tracking behaviour of your system, since most linear controllers are designed to stabilize rest positions.