This answer explained steps to building time series model for forecasting. However, the model is used for one-step ahead. How can I perform multi-step ahead?
1 Answer
I'm not an expert on MLP time series models, but logically there are two ways you can do this. These examples asume you want to skip 2 steps, and the trainingset is x1, x2, x3, xn...
Method 1, teach to skip months:
You should train your network as follows:
Input: [x1, x2, x3], Output: [x6]
You basically just teach it the pattern between xn, xn+1, xn+2
and xn+5
.
Method 2, predict multiple months
This advise is more eye-appealing, but it requires more steps:
Input: [x1, x2, x3], Output: [x4]
-> save x4
to next step
Input: [x2, x3, x4], Output: [x5]
-> save x5
to next step
Input: [x3, x4, x5], Output: [x6]
-> you have your output