Neural Network: What kind of NARX do i need: one step or multi step prediction

Hello,

I'm a beginner in ANN and I want to use it to make predictions of elevation changes over time Z(t). I have measured data of an applied force F(t), which is all the time (past and future) available and can work as input and past data of elevation changes, which can work as the target. I think NARX can work pretty well in my case, as future development of the elevation Z(t+1) depends on the force F(t+1) and the previous elevation Z(t) - as at a certain point more force will not lead to a further development.

• Question 1: Is NARX a good tool for this task, or has someone a better idea?

I studied the provided example NARX-Maglev but I could not really get what is the difference between "Multi Step Ahead prediction" or "one-step ahead prediction" - and which one I need.

• Question 2: Which type of prediction do I need, to make predictions in the future for every timestep?

Many thanks, Markus

Answer 1

[Sachin Meena]

From the information provided, I believe that NARX would be a good fit for your data. As for the type of prediction, you can read the following and choose as per your use case.

One step prediction: This means that F(t+1) will be used to predict only Z(t+1).

Multi step prediction: This means that F(t+1) will be used to predict values of Z(y) where y ranges from [t+1,t+k] as specified.

I believe one step prediction is suitable for your use case.