Neural network for road profile reconstruction

I want to use a neural network to predict road profile elevations from acceleration data. I have many road profiles with measured elevations by steps of 0.1 m. So at each 0.1 m, I know the elevation of the road in mm, for total distances as long as 20 km or more for each profile. I then use, quarter car or full car simulation models to generate accelerations associated to each road profile. My task is to use these accelerations as inputs for the neural network to determine a road profile as output, as close as possible to the original profile (target).

Question 1: What is the most appropriate kind of neural network to use for this problem.

Question 2: I need to feed to the network the training samples. I think it is wise to divide the profiles that are km long in smaller parts and feed these to the networks as training samples. Is it right? If yes, how big these parts should be? 1 m? 10 m? 100 m?