How can i obtain a Narx block in Simulink independent from the sample time of the identified neural network?

Hi everyone,

I've trained a NARX neural network with an input\output timeseries data sampled at dt = 0.00125.

Then, i've converted the net in a closed loop form thanks to the closeloop function.

Finally, i've generated a Simulink block of this closed net by using gensim function and by specifying a sample time equal to the one of the timeseries used for the identification (so, exactly dt = 0.00125).

Now, i want to use this gensim block in another Simulink model, where the sample time is different than dt = 0.00125 and could be any.

My wish would be to create a block that is independent of the sample time and adapts to the sample time of the simulink model simulation.

Answer 1

[David Willingham]

Marco,

As the NNet is trained on a specific sample time, it's likely not going to predict well at different sample times. My recommendation is that you'll need to add a rate transistion block before / after the NNet. An example of how to do this can be found here:

https://www.mathworks.com/help/simulink/ug/how-to-specify-the-sample-time.html

David

[Marco Pizzoli]

Hi David,

Thank you very much for your reply!

I will definitely try the solution you recommended.

Marco