Merge pull request #538 from opendsm/fix/ghi-sufficiency

Add GHI sufficiency check requiring 90% coverage for each month

CHANGELOG.md

@@ -4,7 +4,7 @@ Changelog
 Development
 -----------
 
-* Placeholder
+* Add GHI sufficiency check requiring 90% coverage for each month
 
 4.1.0
 -----

eemeter/eemeter/common/sufficiency_criteria.py

@@ -469,6 +469,25 @@ def _check_hourly_consecutive_temperature_data(self):
                 )
             )
 
+    def _check_monthly_ghi_percentage(self):
+        if "ghi" not in self.data.columns:
+            return
+        non_null_temp_ghi_per_month = (
+            self.data["ghi"]
+            .groupby(self.data.index.month)
+            .apply(lambda x: x.notna().mean())
+        )
+        if (non_null_temp_ghi_per_month < self.min_fraction_daily_coverage).any():
+            self.disqualification.append(
+                EEMeterWarning(
+                    qualified_name="eemeter.sufficiency_criteria.missing_monthly_ghi_data",
+                    description=("At least one month is missing over 10% of GHI data."),
+                    data={
+                        "lowest_monthly_coverage": non_null_temp_ghi_per_month.min(),
+                    },
+                )
+            )
+
     def check_sufficiency_baseline(self):
         # TODO : add caltrack check number on top of each method
         self._check_no_data()
@@ -480,6 +499,7 @@ def check_sufficiency_baseline(self):
         self._check_monthly_temperature_values_percentage()
         self._check_monthly_meter_readings_percentage()
         self._check_extreme_values()
+        self._check_monthly_ghi_percentage()
         # TODO these will only apply to legacy, and currently do not work
         # self._check_high_frequency_meter_values()
         # self._check_high_frequency_temperature_values()
@@ -490,6 +510,7 @@ def check_sufficiency_reporting(self):
         self._check_valid_days_percentage()
         self._check_valid_temperature_values_percentage()
         self._check_monthly_temperature_values_percentage()
+        self._check_monthly_ghi_percentage()
         # self._check_high_frequency_temperature_values()
 
 

eemeter/eemeter/models/hourly/data.py

@@ -417,6 +417,8 @@ def _create_sufficiency_df(df: pd.DataFrame):
     """Creates dataframe equivalent to legacy hourly input"""
     df.loc[df["interpolated_observed"] == 1, "observed"] = np.nan
     df.loc[df["interpolated_temperature"] == 1, "temperature"] = np.nan
+    if "ghi" in df.columns:
+        df.loc[df["interpolated_ghi"] == 1, "ghi"] = np.nan
     # set temperature_not_null  to 1.0 if temperature is not null
     df["temperature_not_null"] = df["temperature"].notnull().astype(float)
     df["temperature_null"] = df["temperature"].isnull().astype(float)

tests/test_hourly_model.py

@@ -80,6 +80,9 @@ def test_good_data(baseline, reporting):
     reporting_data = HourlyReportingData(reporting, is_electricity_data=True)
     hm = HourlyModel().fit(baseline_data)
     p1 = hm.predict(reporting_data)
+    assert np.isclose(
+        p1["predicted"].sum(), 1135000, rtol=1e-2
+    )  # quick check that model fit isn't changing drastically
     serialized = hm.to_json()
     hm2 = HourlyModel.from_json(serialized)
     p2 = hm2.predict(reporting_data)
@@ -259,6 +262,27 @@ def test_monthly_percentage(baseline):
         HourlyModel().fit(baseline_data)
 
 
+def test_monthly_ghi_percentage(baseline):
+    # create datetimeindex where a little over 10% of days are missing in feb, but still 90% overall
+    missing_idx = pd.date_range(
+        start=baseline.index.min(), end=baseline.index.max(), freq="h"
+    )
+    missing_idx = missing_idx[missing_idx.day < 4]
+
+    invalid_ghi = baseline.copy()
+    invalid_ghi.loc[invalid_ghi.index.day < 5, "ghi"] = np.nan
+
+    baseline_data = HourlyBaselineData(invalid_ghi, is_electricity_data=True)
+    assert_dq(
+        baseline_data,
+        [
+            "eemeter.sufficiency_criteria.missing_monthly_ghi_data",
+        ],
+    )
+    with pytest.raises(DataSufficiencyError):
+        HourlyModel().fit(baseline_data)
+
+
 def test_hourly_fit_daily_threshold(baseline):
     """confirm that days with >50% interpolated data are excluded from fit step"""