The thinking being that maybe it's unrealistic to expect the model to learn by rote the differences of each PV system. Instead, the recent history should tell us what we need to know, and also might mean that we can do inference for unseen PV systems without retraining
Possibly instead of the PV system ID embedding
The thinking being that maybe it's unrealistic to expect the model to learn by rote the differences of each PV system. Instead, the recent history should tell us what we need to know, and also might mean that we can do inference for unseen PV systems without retraining