How can I feed the outputs of a predict block back into its inputs?

Hello all,

I would like to return the outputs of a Predict block to its input in Simulink. In the Predict block I use a neural network of type DLNetwork, which is forward directed (not recursive). As input format I use 'CT'. The network expects 5 inputs and gives 2 outputs. If I do not feed back the inputs, everything works. Here is a minimal example:

But when I feed back the inputs, I get the following error messages:

• Layer 'Input': Invalid input data. Invalid size of channel dimension. Layer expects input with channel dimension size 5 but received input with size 1. Function 'Predict1/MLFB' (#1113.64.309), line 2, column 19: "deep.blocks.internal.deepNetwork({in_1}, {size(in_1)}, {class(in_1)}, coder.cons" Launch diagnostic report.
• 'untitled1/Predict1/MLFB' or the model referenced by it contains a block that updates persistent or state variables while computing outputs and is not supported in an algebraic loop. It is in an algebraic loop with the following blocks.

How do I manage in Simulink to return the outputs of the Predict block without getting an error?

Many thanks in advance.

[VBBV]

Use of summer junction for first 2 signals causes the signals to be nullified making it essentially same as without signal.

Delete the summer junction and attach them using a different bus connector

[teasy]

What is a Summer junction? I only use mux and demux blocks. These should be suitable for my operation according to the documentation.

[Purvaja]

Hey @teasy,

Since you are feeding the signals back into the input, the model cannot determine whether the signals are instantaneous or when they should be processed relative to other steps. I recommend using a Unit Delay or Memory block to break the resulting algebraic loop.

I attempted to replicate your model, but encountered several errors. It would be very helpful if you could provide the .slx model file, the neural network, and the specific block parameters you have configured.