How do I train a NARX network with multiple time series datasets?

Hello there,

I am a bit new to neural networks. There is a single outlet from a dam that distributes water into little canals. The flow in the dam outlet 'x(t)' controls the flow in any given canal 'y(t)'. However, the flow in a given canal also depends on some time-independent variable, say x2 = permeability of the canal. Now I have many datasets for many different canals connected to the same outlet, each having the following columns, [flow in dam outlet, flow in given canal, permeability of canal]. The permeability remains the same for a given canal.

I don't know how to feed all these different datasets to the time series. I use the neural network toolbox and it accepts only a single time series. Can anyone please explain how I can train my network so that it can give me the flow in the canal y(t), given that i know the flow in the dam outlet x(t) and the permeability of the canal, x2 (which is independent of time)...

[Luke135]

I am also interested in the answer since I am facing a similar problem.

[Matheus Torquato]

I also have a similar question. I have 200 cycles of data from 100 engines (with 24 parameters each).

I have organized it like this:

{Engine1_Time001}    {Engine2_Time001}            {Engine200_Time001}
{Engine1_Time002}    {Engine2_Time002}   .  .  .  {Engine200_Time002}
{Engine1_Time003}    {Engine2_Time003}            {Engine200_Time003}
 .                    .                  .  
 .                    .                     .
 .                    .                        .
{Engine1_Time200}    {Engine2_Time200}            {Engine200_Time100}

Where {EngineX_TimeY} is a 1x24 Cell containing 24 parameters of the Engine X at cycle Y.

The Neural Network Time Series App doesn't seem to be able to handle that structure since the options available are only Cell Column, Matrix Column and Matrix Row.

How can I train this NARX network?

[Yougzz]

Hello, I have the same problem, did you find a solution ?

Answer 1

[Luke135]

Can this be helpful? https://nl.mathworks.com/help/nnet/ug/multiple-sequences-with-dynamic-neural-networks.html

[Federico Alberto Gorrini]

This is great! Thanks guy!

"Multiple Sequences with Dynamic Neural Networks"