def gaussian(data, wind_farm, inplace=True, curve="state"):
"""Impute missing data using gaussian distributions of U & V. For each
missing entry, sample U & V based on mean and covariance of non-missing
entries that have the same location, same month, and same hour.
:param pandas.DataFrame data: data frame as returned by
:py:func:`prereise.gather.winddata.rap.rap.retrieve_data`.
:param pandas.DataFrame wind_farm: data frame of wind farms.
:param bool inplace: should the imputation be done in place.
:param str curve: 'state' to use the state average, otherwise named curve.
:return: (*pandas.DataFrame*) -- data frame with missing entries imputed.
"""
_check_curve(curve)
data_impute = data if inplace else data.copy()
to_impute = _find_to_impute(data)
if to_impute is None:
return
# Information on wind turbines & state average tubrine curves
tpc = get_turbine_power_curves()
spc = get_state_power_curves()
# Timestamp of all entries in data frame
dates = pd.DatetimeIndex(data.index.values)
n_target = len(wind_farm)
select = None
for i, hour in tqdm(enumerate(to_impute), total=len(to_impute)):
# Only run the similar-selection function the first time
if i % n_target == 0:
select = _select_similar(data, dates, hour)
plant_id = data.loc[hour].plant_id
select_plant = select[select.plant_id == plant_id]
uv_data = np.array(
[select_plant["U"].to_numpy(), select_plant["V"].to_numpy()])
cov = np.cov(uv_data)
mean = np.mean(uv_data, axis=1)
sample = np.random.multivariate_normal(mean=mean, cov=cov, size=1)
data_impute.at[hour, "U"] = sample[0][0]
data_impute.at[hour, "V"] = sample[0][1]
wspd = np.sqrt(data.loc[hour].U**2 + data.loc[hour].V**2)
normalized_power = get_power(tpc, spc, wspd, "IEC class 2")
data_impute.at[hour, "Pout"] = normalized_power
if not inplace:
return data_impute
def simple(data, wind_farm, inplace=True, curve="state"):
"""Impute missing data using a simple procedure. For each missing entry,
the extrema of the U and V components of the wind speed of all non missing
entries that have the same location, same month, same hour are first found
for each missing entry. Then, a U and V value are randomly generated
between the respective derived ranges.
:param pandas.DataFrame data: data frame as returned by
:py:func:`prereise.gather.winddata.rap.rap.retrieve_data`.
:param pandas.DataFrame wind_farm: data frame of wind farms.
:param bool inplace: should the imputation be done in place.
:param str curve: 'state' to use the state average, otherwise named curve.
:return: (*pandas.DataFrame*) -- data frame with missing entries imputed.
"""
_check_curve(curve)
data_impute = data if inplace else data.copy()
to_impute = _find_to_impute(data)
if to_impute is None:
return
# Information on wind turbines & state average tubrine curves
tpc = get_turbine_power_curves()
spc = get_state_power_curves()
# Timestamp of all entries in data frame
dates = pd.DatetimeIndex(data.index.values)
n_target = len(wind_farm)
select = None
for i, j in tqdm(enumerate(to_impute), total=len(to_impute)):
if i % n_target == 0:
select = _select_similar(data, dates, j)
k = data.loc[j].plant_id
select_plant = select[select.plant_id == k]
min_u, max_u = select_plant["U"].min(), select_plant["U"].max()
min_v, max_v = select_plant["V"].min(), select_plant["V"].max()
data_impute.at[j, "U"] = min_u + (max_u - min_u) * np.random.random()
data_impute.at[j, "V"] = min_v + (max_v - min_v) * np.random.random()
wspd = np.sqrt(data.loc[j].U**2 + data.loc[j].V**2)
normalized_power = get_power(tpc, spc, wspd, "IEC class 2")
data_impute.at[j, "Pout"] = normalized_power
if not inplace:
return data_impute
def test_get_state_power_curves(self):
state_power_curves = get_state_power_curves()
self.assertIsInstance(state_power_curves, pd.DataFrame)
self.assertEqual(state_power_curves.index.name, "Speed bin (m/s)")
def setUp(self):
self.tpc = get_turbine_power_curves()
self.spc = get_state_power_curves()
def retrieve_data(wind_farm, start_date="2016-01-01", end_date="2016-12-31"):
"""Retrieve wind speed data from NOAA's server.
:param pandas.DataFrame wind_farm: plant data frame.
:param str start_date: start date.
:param str end_date: end date (inclusive).
:return: (*tuple*) -- First element is a pandas data frame with
*'plant_id'*, *'U'*, *'V'*, *'Pout'*, *'ts'* and *'ts_id'* as columns.
The power output is given for a 1MW generator and the U and V component of
the wind speed 80-m above ground level are in m/s. Second element is a list
of missing files.
"""
# Define query box boundaries using the most northern, southern, eastern
# and western. Add 1deg in each direction
north_box = wind_farm.lat.max() + 1
south_box = wind_farm.lat.min() - 1
west_box = wind_farm.lon.min() - 1
east_box = wind_farm.lon.max() + 1
# Information on wind turbines & state average tubrine curves
tpc = get_turbine_power_curves()
spc = get_state_power_curves()
# Information on wind farms
n_target = len(wind_farm)
lon_target = wind_farm.lon.values
lat_target = wind_farm.lat.values
id_target = wind_farm.index.values
state_target = [
"Offshore" if wind_farm.loc[i].type == "wind_offshore" else
id2abv[wind_farm.loc[i].zone_id] for i in id_target
]
start = datetime.datetime.strptime(start_date, "%Y-%m-%d")
end = datetime.datetime.strptime(end_date, "%Y-%m-%d")
box = {
"north": north_box,
"south": south_box,
"west": west_box,
"east": east_box
}
noaa = NoaaApi(box)
url_count = len(noaa.get_path_list(start, end))
missing = []
target2grid = OrderedDict()
size = url_count * n_target
data = pd.DataFrame({
"plant_id": [0] * size,
"ts": [np.nan] * size,
"ts_id": [0] * size,
"U": [0] * size,
"V": [0] * size,
"Pout": [0] * size,
})
dt = datetime.datetime.strptime(start_date, "%Y-%m-%d")
step = datetime.timedelta(hours=1)
def calc_angular_dist(lon_grid, lat_grid):
n_grid = len(lon_grid)
for j in range(n_target):
uv_target = ll2uv(lon_target[j], lat_target[j])
angle = [
angular_distance(uv_target, ll2uv(lon_grid[k], lat_grid[k]))
for k in range(n_grid)
]
target2grid[id_target[j]] = np.argmin(angle)
def handle_missing(response, data_tmp):
missing.append(response.url)
# missing data are set to NaN.
data_tmp["U"] = [np.nan] * n_target
data_tmp["V"] = [np.nan] * n_target
data_tmp["Pout"] = [np.nan] * n_target
first = True
request_iter = enumerate(noaa.get_hourly_data(start, end))
for i, response in tqdm(request_iter, total=url_count):
data_tmp = pd.DataFrame({
"plant_id": id_target,
"ts": [dt] * n_target,
"ts_id": [i + 1] * n_target
})
if response.status_code == 200:
try:
# see demo notebook to understand file structure
tmp = Dataset("tmp.nc", "r", memory=response.content)
lon_grid = tmp.variables["lon"][:].flatten()
lat_grid = tmp.variables["lat"][:].flatten()
u_wsp = tmp.variables[NoaaApi.var_u][0, 1, :, :].flatten()
v_wsp = tmp.variables[NoaaApi.var_v][0, 1, :, :].flatten()
if first:
# The angular distance is calculated once. The target to grid
# correspondence is stored in a dictionary.
calc_angular_dist(lon_grid, lat_grid)
first = False
data_tmp["U"] = [
u_wsp[target2grid[id_target[j]]] for j in range(n_target)
]
data_tmp["V"] = [
v_wsp[target2grid[id_target[j]]] for j in range(n_target)
]
wspd_target = np.sqrt(
pow(data_tmp["U"], 2) + pow(data_tmp["V"], 2))
power = [
get_power(tpc, spc, wspd_target[j], state_target[j])
for j in range(n_target)
]
data_tmp["Pout"] = power
except Exception:
print(f"Failed to parse response from url={response.url}")
handle_missing(response, data_tmp)
else:
handle_missing(response, data_tmp)
data.iloc[i * n_target:(i + 1) * n_target, :] = data_tmp.values
dt += step
# Format data frame
data["plant_id"] = data["plant_id"].astype(np.int32)
data["ts_id"] = data["ts_id"].astype(np.int32)
data["U"] = data["U"].astype(np.float32)
data["V"] = data["V"].astype(np.float32)
data["Pout"] = data["Pout"].astype(np.float32)
data.sort_values(by=["ts_id", "plant_id"], inplace=True)
data.reset_index(inplace=True, drop=True)
return data, missing
def calculate_pout(wind_farms, start_dt, end_dt, directory):
"""Calculate power output for wind farms based on hrrr data.
Function assumes that user has already called
:meth:`prereise.gather.winddata.hrrr.hrrr.retrieve_data` with the same
start_dt, end_dt, and directory.
:param pandas.DataFrame wind_farms: plant data frame.
:param str start_dt: start date.
:param str end_dt: end date (inclusive).
:param str directory: directory where hrrr data is contained.
:return: (*pandas.Dataframe*) -- Pandas containing power out per wind farm
on a per hourly basis between start_dt and end_dt inclusive. Structure of
dataframe is:
wind_farm1 wind_farm2
dt1 POUT POUT
dt2 POUT POUT
"""
turbine_types = wind_farms.apply(
lambda x: "Offshore" if x["type"] == "wind_offshore" else id2abv[x["zone_id"]],
axis=1,
)
turbine_power_curves = get_turbine_power_curves()
state_power_curves = get_state_power_curves()
wind_data_lat_long = get_wind_data_lat_long(start_dt, directory)
wind_farm_to_closest_wind_grid_indices = find_closest_wind_grids(
wind_farms, wind_data_lat_long
)
dts = pd.date_range(start=start_dt, end=end_dt, freq="H").to_pydatetime()
# Fetch wind speed data for each wind farm (or store NaN as applicable)
wind_speed_data = pd.DataFrame(index=dts, columns=wind_farms.index, dtype=float)
for dt in tqdm(dts):
gribs = pygrib.open(formatted_filename(dt))
try:
u_component = gribs.select(name=U_COMPONENT_SELECTOR)[0].values.flatten()
v_component = gribs.select(name=V_COMPONENT_SELECTOR)[0].values.flatten()
wind_farm_specific_u_component = u_component[
wind_farm_to_closest_wind_grid_indices
]
wind_farm_specific_v_component = v_component[
wind_farm_to_closest_wind_grid_indices
]
wind_speed_data.loc[dt] = np.sqrt(
pow(wind_farm_specific_u_component, 2)
+ pow(wind_farm_specific_v_component, 2)
)
except ValueError:
# If the GRIB file is empty, no wind speed values can be selected
wind_speed_data.loc[dt] = np.nan
# For each column, linearly interpolate any NaN values
linear(wind_speed_data)
# Then calculate wind power based on wind speed
wind_power_data = [
[
get_power(
turbine_power_curves,
state_power_curves,
wind_speed_data.loc[dt, w],
turbine_types.loc[w],
)
for w in wind_farms.index
]
for dt in tqdm(dts)
]
df = pd.DataFrame(data=wind_power_data, index=dts, columns=wind_farms.index)
return df