In the data loader: For each 5-minute timestep, pre-compute the GSP time index where gsp_30m_timestep == pv_5m_timestep.ceil("30T"). Then create two new tensors: gsp_5_min of the GSP power at that timestep, and another with the GSP fourier features at that timestep (gsp_5_min_time_utc). (Which, if we implement it as an xr_batch_processor then I think we just need to end the key names with time_utc to get fourier encoding).
In the data loader: For each 5-minute timestep, pre-compute the GSP time index where
gsp_30m_timestep == pv_5m_timestep.ceil("30T")
. Then create two new tensors:gsp_5_min
of the GSP power at that timestep, and another with the GSP fourier features at that timestep (gsp_5_min_time_utc
). (Which, if we implement it as anxr_batch_processor
then I think we just need to end the key names withtime_utc
to get fourier encoding).