✨ Add PV yield prediction via predict_pv_yield_nwp

[tomwhite]

This is not yet ready for merge, but it shows the idea behind #33. It still needs changing to use UKV irradiance data.

cc @JackKelly @fwirtz

Fixes #33

[tomwhite]

Updated to get irradiance data from a Met Office UKV netcdf file. This should be easy to change to read from a Zarr store.

The output is JSON that looks like this:

[
    {
        "system_id":973,
        "easting":445587,
        "northing":497235,
        "time":1591290000000,
        "pv_yield_predicted":507.5
    },
    {
        "system_id":1364,
        "easting":431430,
        "northing":272536,
        "time":1591290000000,
        "pv_yield_predicted":214.8
    },
...
]

[flowirtz]

Awesome, thanks @tomwhite!

Just for my (very limited) understanding: Did you use the PV data from GCP to train the model once, and now for the inference bit we only rely on the NWP Metoffice data? Until we retrain, I suppose.

The output already looks usable, thanks. I'd prefer to use lat/long over easting/northing though, to future proof a bit for expanding to other countries as well. Also, it would be best if the output was GeoJSON.

I think the shape would look something like this then:

{
  "type": "Feature",
  "geometry": {
    "type": "Point",
    "coordinates": [-1.299849, 54.368597]
  },
  "properties": {
    "system_id": 973,
    "time": 1591290000000,
    "pv_yield_predicted": 507.5
  }
}

But you can also keep the shape this way, I'd just convert it in the FE then.
Thanks so much!!

[tomwhite]

Did you use the PV data from GCP to train the model once, and now for the inference bit we only rely on the NWP Metoffice data? Until we retrain, I suppose.

Yes that's right.

Also, it would be best if the output was GeoJSON.

Great idea! I will try to do that on the next iteration of this PR.

[tomwhite]

I've updated this to produce GeoJSON, e.g.

{
    "type": "FeatureCollection",
    "features": [
        {
            "type": "Feature",
            "geometry": {
                "type": "Point",
                "coordinates": [
                    -1.299834,
                    54.3686
                ]
            },
            "properties": {
                "system_id": 973,
                "time": "2020-06-04T17:00:00",
                "pv_yield_predicted": 507.5329670623049
            }
        },
        {
            "type": "Feature",
            "geometry": {
                "type": "Point",
                "coordinates": [
                    -1.54,
                    52.35
                ]
            },
            "properties": {
                "system_id": 1364,
                "time": "2020-06-04T17:00:00",
                "pv_yield_predicted": 214.8320996361009
            }
        },
...
   ]
}

[tomwhite]

This could be merged now and the change to use irradiance data done in a separate PR, or we could keep this open for that. Happy to do it either way.

[flowirtz]

Awesome! Thanks Tom!

[flowirtz]

Merging now, raised #41 to track the remaining work.