How to implement SeriesNetwork in Simulink (Deep Learning)

I followed this youtube video to implement a neural network trained using trainNetwork in simulink.

The neural network takes in a 1x13 vector and ouputs a 1x13 vector (It is meant to be an autoencoder). There is a zero hold block with a sample time of 20s.

Following the predict block method, I got this size mismatch error.

Following the MATLAB function implementation,

function reg = output(u)

persistent network;
if isempty(network)
    network = coder.loadDeepLearningNetwork('autoenc.mat','myautoenc');
end

reg = network.predict(u');
end

I get this error:

How do I fix this?

Answer 1

[Krishna]

Hey Patrick,

Given that you're working on modeling a series network, I'm under the impression you're employing a Recurrent Network. It might be more effective to utilize a stateful predict rather than the regular predict block. This is because stateful predict blocks are specifically designed to handle sequence information in the time domain, which is a key characteristic of recurrent networks.

Please go through the following documentation to learn more,

https://in.mathworks.com/help/deeplearning/ref/statefulpredict.html

Hope this helps.