Why are predicted outputs different between Simulink and Matlab?

I saved trained dlnetwork as "net.mat" before this code.

load("net.mat","net_m");
open_system("simulink");
out = sim("simulink.slx");
x = dlarray(out.simout.Data,"TC");
y = predict(net_m,x);
t  = [0:0.1:10];
figure;
plot(t,extractdata(y));
hold on
plot(t,extractdata(x))
hold off

With above code, the prediction was like blue line in 1st picture.

But in Simulink, prediction was like blue line in 2nd picture.

I'm sure that the filepath in predict block was "net.mat."

Answer 1

[Ben]

Your network is a 1D CNN over the sequence. Simulink executes this network 1 time step at a time. To compare:

x = dlarray(rand(1,100),"CT");

y = predict(net_m,x)

This computes the 1D CNN over the sequence x. However in your Simulink system, you are only doing:

for i = 1:100

y(i) = predict(net_m,x(i));

end

To get the behaviour in Simulink you will need to create a buffer of the sequence/signal input to the Predict block. It might be possible to do this with a Tapped Delay block or a Buffer block. You may need to set the Simulink solver to a fixed step solver with appropriate time step.