Range output = [10 , 6 800 000]. In this range network won't be (or bad) learn.
Scale the output range. For sample. Range output = [10 6,800,000]/10 000 000=[1.0000e-006 , 0.68].
After train or adapt.
Your array = sim (input)* 10 000 000
And set new value for learning rate, goal... etc.