Solarpower Predicting with Sky Images

Dear Forum,

I’m facing some challenges and need your help.

I built a computer controlled all sky imager. I would like to label every 30 s a hermispherical cloud image with the current Solarpower intentisity [W/m²].

I)So would it be possible to generate a trainings dataset which is already labeled, since I get image and label value simultanously every day. The first try will be with 30days of data ~ 15.000 Images

II) Which type of neuronal network (architecture/type) is most for my use case?