add function to run forecast

quartz_solar_forecast/eval/forecast.py

@@ -0,0 +1,102 @@
+import os
+import pandas as pd
+import xarray as xr
+from psp.data_sources.nwp import NwpDataSource
+from psp.data_sources.pv import NetcdfPvDataSource
+from psp.serialization import load_model
+from psp.typings import X
+
+from quartz_solar_forecast.data import get_nwp, make_pv_data
+from quartz_solar_forecast.pydantic_models import PVSite
+
+from datetime import datetime
+
+dir_path = os.path.dirname(os.path.realpath(__file__))
+
+
+def run_forecast(pv_df: pd.DataFrame, nwp_df: pd.DataFrame, nwp_source="ICON") -> pd.DataFrame:
+    """
+    Run the forecast from NWP data
+
+    :param pv_df: the PV site data. This should have columns timestamp, id, latitude, longitude, and capacity
+    :param nwp_df: all the nwp data for the site and location. This shoulw have the following rows
+        - timestamp: the timestamp of the site
+        - temperature_2m
+        - precipitation
+        - shortwave_radiation
+        - direct_radiation",
+        - cloudcover_low",
+        - cloudcover_mid",
+        - cloudcover_high",
+        maybe more
+    """
+
+    # load model only once
+    model = load_model(f"{dir_path}/../models/model-0.3.0.pkl")
+
+    all_predictions = []
+    for i in range(len(pv_df)):
+
+        pv_row = pv_df.iloc[i]
+
+        site = PVSite(
+            latitude=pv_row["latitude"],
+            longitude=pv_row["longitude"],
+            capacity_kwp=pv_row["capacity"],
+        )
+
+        nwp_site_df = nwp_df[
+            (nwp_df["id"] == pv_row.pv_id) & (nwp_df["timestamp"] == pv_row.timestamp)
+        ]
+
+        pv_id = pv_df["pv_id"][i]
+        ts = pv_df["timestamp"][i]
+
+        # format
+        times = nwp_site_df["time"]
+        step = times - ts
+        nwp_site_df = nwp_site_df.drop(columns=["id", "timestamp"])
+        nwp_site_df.set_index("time", inplace=True, drop=True)
+
+        if isinstance(ts, str):
+            ts = datetime.fromisoformat(ts)
+
+        # make pv and nwp data from GFS
+        nwp_xr = xr.DataArray(
+            data=nwp_site_df.values,
+            dims=["step", "variable"],
+            coords=dict(
+                step=("step", step),
+                variable=nwp_site_df.columns,
+            ),
+        )
+        nwp_xr = nwp_xr.to_dataset(name=nwp_source)
+        nwp_xr = nwp_xr.assign_coords(
+            {"x": [site.longitude], "y": [site.latitude], "time": [nwp_site_df.index[0]]}
+        )
+
+        pv_xr = make_pv_data(site=site, ts=ts)
+
+        # format pv and nwp data
+        pv_data_source = NetcdfPvDataSource(
+            pv_xr,
+            id_dim_name="pv_id",
+            timestamp_dim_name="timestamp",
+            rename={"generation_wh": "power", "kwp": "capacity"},
+            ignore_pv_ids=[],
+        )
+        # make NwpDataSource
+        nwp = NwpDataSource(paths_or_data=nwp_xr, value_name=nwp_source)
+        model.set_data_sources(pv_data_source=pv_data_source, nwp_data_sources={nwp_source: nwp})
+
+        # make prediction # TODO change '1'
+        x = X(pv_id="1", ts=ts)
+        pred = model.predict(x)
+
+        # format into timerange and put into pd dataframe
+        times = pd.date_range(start=x.ts, periods=len(pred.powers), freq="15min")
+        pred_df = pd.DataFrame({"power_wh": pred.powers}, index=times)
+
+        all_predictions.append(pred_df)
+
+    return pd.concat(all_predictions)

quartz_solar_forecast/evaluation.py

@@ -6,6 +6,7 @@
 
 """
 from quartz_solar_forecast.eval.nwp import get_nwp
+from quartz_solar_forecast.eval.forecast import run_forecast
 
 import pandas as pd
 
@@ -22,6 +23,8 @@ def run_eval(testset_path):
     nwp_df = get_nwp(testset)
 
     # Run forecast with PV and NWP inputs.
+    # TODO updatepv_df
+    predictions_df = run_forecast(pv_df=None, nwp_df=nwp_df)
 
     # Combine the forecast results with the ground truth (ts, id, horizon (in hours), pred, truth, diff)
 

tests/eval/test_eval_forecast.py

@@ -0,0 +1,71 @@
+from quartz_solar_forecast.eval.forecast import run_forecast
+
+import pandas as pd
+
+
+def test_run_forecast():
+    pv_df = pd.DataFrame(
+        [
+            {
+                "timestamp": pd.Timestamp("2021-01-26 01:15:00"),
+                "latitude": 51.5,
+                "longitude": 0.0,
+                "capacity": 1.0,
+                "pv_id": 1.0,
+            }
+        ]
+    )
+
+    nwp_df = pd.DataFrame(
+        columns=[
+            "id",
+            "timestamp",
+            "t",
+            "prate",
+            "dswrf",
+            "dlwrf",
+            "lcc",
+            "mcc",
+            "hcc",
+            "vis",
+            "si10",
+            "time",
+        ],
+        data=[
+            [
+                1,
+                pd.Timestamp("2021-01-26 01:15:00"),
+                1.0,
+                1.0,
+                1.0,
+                1.0,
+                0.0,
+                0.0,
+                0.0,
+                0,
+                0,
+                pd.Timestamp("2021-01-26 01:15:00"),
+            ],
+            [
+                1,
+                pd.Timestamp("2021-01-26 01:15:00"),
+                1.0,
+                1.0,
+                1.0,
+                1.0,
+                0.0,
+                0.0,
+                0.0,
+                0,
+                0,
+                pd.Timestamp("2021-01-26 01:30:00"),
+            ],
+        ],
+    )
+
+    # format timestamp
+    pv_df["timestamp"] = pd.to_datetime(pv_df["timestamp"])
+    nwp_df["timestamp"] = pd.to_datetime(nwp_df["timestamp"])
+    nwp_df["time"] = pd.to_datetime(nwp_df["time"])
+
+    _ = run_forecast(pv_df=pv_df, nwp_df=nwp_df, nwp_source="ICON")