# Feeding multiple time-series into a neural net - configuration?

I've been looking at how to use neural nets for time-series analysis. The examples I've seen mostly rely on a sliding window to feed the slice of the data into the input layer - which has as many input neurons as there are elements within that window.

Suppose I wanted to feed two "parallel" time-series into the same network - ie. I would like to make a prediction based not just on one time-series, but two time-synchronized series - how should I configure the inputs?

My gut feeling is that I should just give the input layer twice as many inputs to accommodate two windows, one from each series, and let the network sort things out by itself. is this correct? Or should I configure the input layer some other way? Thanks.

From my brief understanding on the subject, I believe you should add the extra time series in the following manner: Say you want to predict $X = (x_0,...,x_t)$. What you say you are doing with sliding window is correct thus start by

Features | target

$x_{t-4}, x_{t-3} | x_{t-2}$

$x_{t-3}, x_{t-2} |x_{t-1}$

$x_{t-2}, x_{t-1} | x_t$

But you also want another time series used as input, say $Z=(z_0,...,z_t)$ so I would imagine you should do it like this:

Features | target

$x_{t-4}, x_{t-3}, z_{t-3} | x_{t-2}$

$x_{t-3}, x_{t-2}, z_{t-2} |x_{t-1}$

$x_{t-2}, x_{t-1}, z_{t-1} | x_t$