I think that topography data you are hoping to work from is too coarse, and the timing of the lights and other obstructions is impossible to get a handle on.
I do however think a vehicle equipped with a handful of sensors could be driven over the course and allow more accurate predictions about how to drive that specific course next time, but the data generally available online is not accurate enough to lend itself to accurate throttle position recommendations.
I did like the hill testing approach though. Adding in powered acceleration introduces enough additional variables that it is worth it to make sure the model is accurate without it. But you need a detailed profile of said hill, which at this point looks like it would take a tape measure and an inclinometer
