You need to return to the fundamentals of curve fitting. Curve fitting is the art of minimizing the error between a set of experimental data and a theoretical model. The key point is "minimizing the error". What you wish to do is take your equation and minimize the distance between the points you generate and the points you have experimentally determined. It is up to you whether this error is error in the V dimension, error in the I dimension, or error in both. In any case, you can create an error function. I would strongly recommend you use a linear (error proportional to linear distance) or logarithmic error (error proportional to log of distance) model rather than a squared model (error proportional to square of distance - used by least squares routines). Technically, these correspond to a double exponetial or lorentz error distribution instead of a gaussian error distribution. They are much more tolerant of experimental errors.
Now that you have an error function, you need a way to minimize it. LabVIEW has several minimizers. I would recommend the downhill simplex, since it almost always works. You should rerun it when you finish, however, to make sure you didn't get hung up in a local minimum.
Please read the section in
Numerical Recipes on robust curve fitting to fill in the details. I can't explain it anywhere near as well as the book does.
