predict_pv_yield issues

https://medium.com/@Cambridge_Spark/coordconv-layer-deep-learning-e02d728c2311 Suggested by @thecapeador The idea being that it's really important for the CNN to discriminate between based on their location. Several combinations to try: - [ ] Location relative...

JackKelly

machine learning model

Use PV metadata (tilt angle etc.)

2

Two approaches: 1) One network which gets PV metadata when it's available. When it's not available, somehow mask those inputs. Set to -1? Or have a separate 'mask' input? 2)...

JackKelly

data

Ongoing thread: Research results & design ideas

2

Keeping track of some basic research results: ## Inferring PV yield for t0 ("now") (not predicting the future): Getting about 6% normalised MAE where the network input is 128x128 pixels...

JackKelly

machine learning model

Dropout PV system ID & PV metadata

During training: With some probability (30%?), zero-out the output from the PV system ID encoding. And, separately, zero-out different elements of the PV metadata. Then, during inference, just feed the...

JackKelly

machine learning model

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...

JackKelly

data