How to predict multivariate time series in Keras?

Question

Neural networks

There is a machine from which information about the value of several sensors is collected in real time. The result is a multidimensional series. And how to work with such, everywhere there are examples with only one-dimensional series?

Answer 1

Well, not exactly everywhere.
For example - https://www.analyticsvidhya.com/blog/2018/09/multi...
And here - using Keras
https://machinelearningmastery.com/multivariate-ti...

Answer 2

I understand that this is only possible in LSTM and similar structures.
You create a three-dimensional array data(x, y, z) where -
x is the number of training samples
y is the number of vectors in one training sample
z is the amount of data in one vector.
Then you create a layer LSTM
model.add(LSTM(units=32, input_shape=(y, z)))
x you don't need to specify the network itself will understand what's what.
units is the number of exits to the next layer (32 as an example).
If my idea is wrong, then I think they will correct me, but it seems to work.