Windnet: Forecast correct values?

peterdudfield commented 6 months ago

Describe the bug

I'm not totalyl convience the forecast is working proerply

Expected behavior

follow the generation values

Additional context

might need to check the batch values are going in correctly
might need to check the live ecmwf values are simiarl to the backtest ones. Sometimes they could be out by a factor of 1000 due to units
are the NWP variables in the correct order?
I had to rename t->t2m and tlc to tcc, is this right? linked with https://github.com/openclimatefix/nwp-consumer/issues/131

Things done / TODO

[x] Updated location in database
[x] Check variables names are ok
[x] TCC looks right, so ok to rename from clt
[x] Check nwp are in the right order, yes ocf_datapipes does this. the xr_data.sel(...) does this
[x] update lon/ lat position
[x] Check normalize variables are the same
[x] Check live and backtest ecmwf are in 0.1 degrees
[x] Check wind speed are similar to open meteo
[ ] compare backtest batch with live batch
[x] Checked that NWP targettimes do change, when running app at different times
[ ] ...

peterdudfield commented 5 months ago

Ive set the centers to be (wind on bottom (EDIT))

@jacobbieker can you confirm?

jacobbieker commented 5 months ago

The latitude is correct for both. The longitude should be wind: 72.6399 pv: 74.6399

peterdudfield commented 5 months ago

Thanks, sorry wind was on the bottom. Interesting, ok I will try locally and check that the correct grid is used. Thats the most important thing right?

jacobbieker commented 5 months ago

Yeah, the grid is the important one, we don't directly pass the lat longs to the model, so it only matters that the grids match

dfulu commented 5 months ago

I've been looking into the forecast performance. Here's a graph of the results including the model version number at the bottom. This is the rolling 15 minute horizon and across the month it got an MAE 240 MW (i.e. 7.11% capacity). I've set the limits on the y-axis on the top plot so the highest value is the capacity (currently 3,381,000 KW).

The forecast seems to have been bad over version 0.1.36?

download (55)

dfulu commented 5 months ago

This is the same graph but for 24 hour horizon

To me it looks like things have been okay since v0.1.37

download (56)

openclimatefix / india-forecast-app