Saving vectors for consecutive timesteps and turn them into an array

I have built a Simulink model which has 5 input signals. Once the scalar values get normalized I turn these 5 signals into a vector by using the Mux Block.

My LSTM (Block: Stateful Predict) needs a numeric array as an input. This works perfectly fine when I am using the constant block with predefined values. However it would not accept a vector as it needs a timeseries.

What would be the best way to save 10 vectors from 10 consecutive timesteps and turn them into an array in order to give my neural network a valid input? I did not find suitable Simulink blocks, so maybe it can only work with a Matlab function block?

Thanks a lot in advance!

Kind Regards

Answer 1

[Samay Sagar]

To provide a sequence of vectors as input to the LSTM 'Stateful Predict' block in Simulink, you can accumulate the vectors over time within a MATLAB Function block and transform it to a “timeseries”.

Here’s how you can implement this:

Add a MATLAB Function block to your Simulink model to store and update a buffer that accumulates vectors from consecutive timesteps. You can now use this output as input to your “Stateful Predict” block.

function timeseriesData = inputToTimeSeries(u)
    % u: Current input vector from the Mux block
    % Define persistent variables for the buffer vector
    persistent buffer;
    buffer = [buffer ;u];
    % Convert the buffer to a timetable
    timeseriesData = array2timetable(buffer, 'TimeStep', seconds(0.2));
end

Read more about “array2timetable” here:

https://www.mathworks.com/help/releases/R2021a/matlab/ref/array2timetable.html