Maybe this might help.
In the 70's, before big data became a practice (because of lack of computing resources),
I was in production seeking answers why yields were not as good as one might expect.
Looking back I started by collecting data, doing simple averaging, and upon inspecting
those simple results that would lead to broadening the data inputs, as well as modifying
its analysis. That was done first for wafer fab, then test, then packaging, and some remarkable,
but not rocket science, results and insights occurred.
This activity resulted in several $M additional revenue over the short term, much more
over the long term. Not chump change in the 70's.
So it seems from your stated goal what you are seeking is the impact of the total heat
load over the day. Which is nothing more than integrating all the samples of T for the day,
eg. the total energy. Rather than averaging which is seeking the "central value" of the
dataset. From integration and observation, you might be able to see some correlation
with a part of the dataset (day) that should be explored further. Or characteristics.
Then keep modifying your measurements/algorithms as you get closer to the meaning
that you seek, even though that may be ill defined at the start.
You might add Humidity and Light Level and barometric pressure to the dataset sensors
as well.
Just some thoughts.....
Regards, Dana.