I have something similar:
I have a string inverter with 2 x 5 panels in series on one mppt input and another 10 panels on the second mppt input.
On one row of these 5 panels I have an item in the way that casts a shadow on one of the panels and an hour later covers the next panel and so on
Dark blue is AC power to the grid, green is MPPT input with the unshaded panels and light blue are the 10 panels of which 5 get a shade during the day.
My dips are certainly not as deep as yours, but you should see if you can find something (air vent, antenna,chimney) that during those times might be casting a shadow on some of your panels.
And like lights in a Christmas tree, if one of the panels gets shaded, the whole strings produces way less (unless you use optimizers underneath each panel)
Most inverters have an MPPT scan time cycle and reset parameter. On mine this is adjustable. At the set time the inverter will reset the MPPT to maximize the Maximum Power Point Tracking. Lets say it is cloudy, the MPPT locks in. But the sun comes out full blast, when the cycle scan comes up it will lock into the brighter sun situation. Down side is the lower you set the scan time, then the more off line you are. But it would adapt to big billowy clouds better. Yours may not be settable. Take a look at the production before the scan cycle and then right after. Usually the solar production will be higher.
As a side note, intermittent and partial shade is one of the big problems for more widespread solar usage. While not yet for rooftop solar, I did notice these guys the other day - they claim a much more shade tolerant solar solution, though mostly for off grid-applications.