Jack Kelly issues

Results 285 issues of


Jack Kelly

Give self-attention raw, unprojected satellite imagery

One of the really cool things about using self-attention is that input images no longer have to confirm to a regular grid. Transformers view their inputs as a set. So...

data

Input into ML model recent PV data from all PV systems in the region of interest

data

When using a transformer, try learning-rate warm up and/or layer norm inside the residual blocks

https://twitter.com/sytelus/status/1411607820542218245

machine learning model

Learning rate decay

machine learning model

Baseline: Predict PV from NWPs a few 10s of km away (to mimic ESO's solar forecasts)

performance analysis: baselines & metrics

Try: no optical flow. Just last n satellite images as a stack of channels. And network predicts next few hours of PV.

One way to approach #30

enhancement

data

machine learning model

Compute validation error per PV system and per time of day

Maybe we're struggling at dawn and dusk? And let's identify bad PV systems

enhancement

performance analysis: baselines & metrics

Find a good metric for measuring if the timing of the forecast is correct

Correlation? Accuracy / f1-score / precision / recall (after thresholding the forecast). Maybe a multi-class classification, motivated by the idea that the grid control room users care a lot about...

enhancement

performance analysis: baselines & metrics

Simple "climatology" baseline: Neural net doesn't get sat image or weather forecasts. Just gets hour of day, day of year, geolocation, clearsky, and Solar azimuth and angle.

enhancement

performance analysis: baselines & metrics

Baseline: Predict PV from just NWP

Irradiance, temperature & wind speed. Try with and without historical PV data (from site of interest, and from neighbouring sites). Could go all-out and use Temporal Fusion Transformer, given the...

enhancement

performance analysis: baselines & metrics