How to judge the relatively stable section in the curve

This is a curve. The relatively stable part in the red box is what I need. How can I automatically judge and recognize it.Can you give a simple automatic judgment code?

[Matt J]

If you only have the one curve, why would you need to do it automatically?

If you have more than one such curve, show us examples of more curves. What is the minimum width a stable region can be? Will there always be only one contiguous stable interval? How large can the oscillations be to be considered stable?

[John D'Errico]

Good questions by Matt. To them, I would add, if this is a reasonable sample from your family of curves, it seems the flat part is also the most noisy. Is that commonly the case? If so, then just look for the region with a high local variance.

The point of all of these questions is you surely know information about this process that you have not told us. But if you want help, then we need to understand what it is that you know, but have kept a secret.

[Jon]

Here is a whole paper on the topic, or at least a closely related one https://www.r3eda.com/wp-content/uploads/2019/05/r3eda-site-SSTSID-Tutorial-2019-05-08.pdf

[Jon]

And here is another, which also seem to closely address the issue you have

https://www.sciencedirect.com/science/article/pii/S2468013320300103

Answer 1

[Matt J]

One possibility might be to look at the local span seminorm (max-min) over a sliding window.

window=50;
lookahead  = movmax(curve,0,window)-movmin(curve,0,window)<=tolerance;
lookbehind = movmax(curve,window,0)-movmin(curve,window,0)<=tolerance;

start=find(lookahead,1,'first');  %start of stable interval
stop=find(lookbehind,1,'last');  %end of stable interval.