issues
search
openclimatefix
/
power_perceiver
Machine learning experiments using the Perceiver IO model to forecast the electricity system (starting with solar)
MIT License
7
stars
1
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
[META] Get Power Perceiver ready for production
#149
JackKelly
closed
2 years ago
1
[META] Improve Helios production model
#148
JackKelly
opened
2 years ago
0
Weight the first hour more highly
#147
JackKelly
closed
2 years ago
1
Explicitly predict national PV power
#146
JackKelly
opened
2 years ago
0
More NWP variables
#145
JackKelly
closed
2 years ago
2
Reduce duplication between `PVQueryGenerator` and `GSPQueryGenerator`
#144
JackKelly
opened
2 years ago
0
Resample NWP to half hourly
#143
JackKelly
closed
2 years ago
0
Modify the NWP data loader so it uses most recent init time for history.
#142
JackKelly
closed
2 years ago
1
The t0_idx should be specified by the DataLoader, and modified by anything that resamples, and passed into the model.
#141
JackKelly
closed
2 years ago
0
always downsample PV to half hourly to keep things simple for now?
#140
JackKelly
closed
2 years ago
0
Implement tweaked model which predicts single timeseries at once
#139
JackKelly
opened
2 years ago
0
Fix bug when trying to predict 8 hrs into the future.
#138
JackKelly
closed
2 years ago
2
Re-implement SunPosition as a RawDataSource
#137
JackKelly
closed
2 years ago
2
If we're not training the u-net, then maybe don't include future satellite imagery in the batches?
#136
JackKelly
opened
2 years ago
1
Try removing the RNN?
#135
JackKelly
closed
2 years ago
2
Give the model all of yesterday's PV & GSP & NWP history
#134
JackKelly
opened
2 years ago
1
[META] forecast 8 hrs into the future
#133
JackKelly
closed
2 years ago
7
Use different random number generator seeds for each GPU process
#132
JackKelly
closed
2 years ago
0
Simplify the GSP queries
#131
JackKelly
opened
2 years ago
0
When training on multiple GPUs, `LogNationalPV` will only see half the GSPs, I think?
#130
JackKelly
closed
2 years ago
2
Use uint8 for satellite imagery?
#129
JackKelly
closed
2 years ago
1
Fine-tune on GSP only
#128
JackKelly
opened
2 years ago
0
Have a `data_source_combo` which selects locations centered on PV systems, not GSPs
#127
JackKelly
opened
2 years ago
0
Train just the u-net on its own on GCP for a few days
#126
JackKelly
opened
2 years ago
0
Maybe we shouldn't train the unet when training the rest of the model
#125
JackKelly
closed
2 years ago
3
More PV systems per example
#124
JackKelly
opened
2 years ago
0
Concurrently load next set of data into RAM whilst training
#123
JackKelly
opened
2 years ago
0
META: Speed up training
#122
JackKelly
opened
2 years ago
0
`detach` the SatellitePredictor after it goes into downstream processing.
#121
JackKelly
closed
2 years ago
0
Add noise to "updated" GSP PV_Live so it looks more like intra-day PV_Live, and train model
#120
JackKelly
opened
2 years ago
0
Use historical GSP PV Live
#119
JackKelly
closed
2 years ago
0
Train on PVoutput.org
#118
JackKelly
opened
2 years ago
0
For test, use the same timestamps as for the v16 test set
#117
JackKelly
opened
2 years ago
0
Always save the most recent model to disk, so we can restart training
#116
JackKelly
closed
2 years ago
0
Plot NWPs in wandb
#115
JackKelly
closed
2 years ago
0
Give the model more like 7 days of PV history
#114
JackKelly
opened
2 years ago
1
Why is the last PV prob plot blank?
#113
JackKelly
closed
2 years ago
0
Aux output from PV encoder RNN
#112
JackKelly
opened
2 years ago
0
Train on multiple GPUs
#111
JackKelly
closed
2 years ago
2
Rename this repo!
#110
JackKelly
opened
2 years ago
0
Give the model the NWP `step`
#109
JackKelly
closed
2 years ago
0
Predict many hours into the future
#108
JackKelly
closed
2 years ago
0
Give the `SatelliteTransformer` air temp and wind speed for that timestep (maybe do that last because that could be a tiny bit fiddly)
#107
JackKelly
opened
2 years ago
0
Can we load the weights for 25.04, and save _just_ the SatellitePredictor's weights, for re-use in other models?
#106
JackKelly
closed
2 years ago
2
Include the Satellite Transformer output in the objective function
#105
JackKelly
closed
2 years ago
0
Try giving PV ID embedding to the PV history encoder RNN
#104
JackKelly
opened
2 years ago
0
Maybe don't mask time_transformer?
#103
JackKelly
opened
2 years ago
0
Try using an RNN for GSP power
#102
JackKelly
opened
2 years ago
0
NWP
#101
JackKelly
closed
2 years ago
0
Validate across national PV.
#100
JackKelly
closed
2 years ago
0
Previous
Next