You don't give your actual data, just a pcture of it. Yes, sometimes a picture is worth a thousand words, but pictures are not worth a thousand numbers. :) A fellow named Mark may argue that point with me. :)
If I wanted to do this, I'd probably use a scheme where I would compute a moving estimate of the slope. Essentially, a variation of Savitsky-Golay filter would do. So at each point in the time series, I would implicitly fit a linear polynomial through the curve. Now I would look for the regions where the slope is uniformly small over a region. Then compute the mean over that interval.
No data, so no way to show you how to do it on your data. Yes. I could make up some data. Do I really want to think? Nah.
