CityBEM-CityLayers-SaeedRay.../hub/imports/weather/epw_weather_parameters.py

160 lines
9.1 KiB
Python

"""
EpwWeatherParameters class to extract weather parameters from a defined region in .epw format (EnergyPlus Weather)
SPDX - License - Identifier: LGPL - 3.0 - or -later
Copyright © 2022 Concordia CERC group
Project Coder Pilar Monsalvete Alvarez de Uribarri pilar.monsalvete@concordia.ca
"""
import logging
import sys
from pathlib import Path
import requests
import pandas as pd
import hub.helpers.constants as cte
from hub.helpers.monthly_values import MonthlyValues
from hub.city_model_structure.city import City
from hub.imports.weather.helpers.weather import Weather as wh
class EpwWeatherParameters:
"""
EpwWeatherParameters class
"""
def __init__(self, city: City):
self._weather_values = None
self._city = city
self._url = wh().epw_file(city.region_code)
self._path = (Path(__file__).parent.parent.parent / f'data/weather/epw/{self._url.rsplit("/", 1)[1]}').resolve()
if not self._path.exists():
with open(self._path, 'wb') as epw_file:
epw_file.write(requests.get(self._url, allow_redirects=True).content)
try:
with open(self._path, 'r', encoding='utf8') as file:
line = file.readline().split(',')
city.climate_reference_city = line[1]
city.latitude = line[6]
city.longitude = line[7]
city.time_zone = line[8]
for i in range(0, 2):
_ = file.readline().split(',')
line = file.readline().split(',')
number_records = int(line[1])
ground_temperature_from_file = {}
for i in range(0, number_records):
depth_measurement_ground_temperature = line[i*16+2]
temperatures = []
for j in range(0, 12):
temperatures.append(float(line[i*16+j+6]))
ground_temperature_from_file[depth_measurement_ground_temperature] = temperatures
except SystemExit:
logging.error('Error: weather file %s not found. Please download it from https://energyplus.net/weather and place'
' it in folder data\\weather\\epw', self._path)
sys.exit()
try:
self._weather_values = pd.read_csv(self._path, header=0, skiprows=7,
names=['year', 'month', 'day', 'hour', 'minute',
'data_source_and_uncertainty_flags',
'dry_bulb_temperature_c', 'dew_point_temperature_c',
'relative_humidity_perc',
'atmospheric_station_pressure_pa',
'extraterrestrial_horizontal_radiation_wh_m2',
'extraterrestrial_direct_normal_radiation_wh_m2',
'horizontal_infrared_radiation_intensity_wh_m2',
'global_horizontal_radiation_wh_m2', 'direct_normal_radiation_wh_m2',
'diffuse_horizontal_radiation_wh_m2',
'global_horizontal_illuminance_lux',
'direct_normal_illuminance_lux', 'diffuse_horizontal_illuminance_lux',
'zenith_luminance_cd_m2',
'wind_direction_deg', 'wind_speed_m_s', 'total_sky_cover',
'opaque_sky_cover', 'visibility_km',
'ceiling_heigh_m_s', 'present_weather_observation',
'present_weather_codes',
'precipitable_water_mm', 'aerosol_optical_depth_10_3_ths',
'snow_depth_cm',
'days_since_last_snowfall', 'albedo', 'liquid_precipitation_depth_mm',
'liquid_precipitation_quality_hr'])
number_invalid_records = self._weather_values[
self._weather_values.dry_bulb_temperature_c == 99.9].count().dry_bulb_temperature_c
if number_invalid_records > 0:
sys.stderr.write(f'Warning: {self._path} invalid records (value of 99.9) in dry bulb temperature\n')
number_invalid_records = self._weather_values[
self._weather_values.global_horizontal_radiation_wh_m2 == 9999].count().global_horizontal_radiation_wh_m2
if number_invalid_records > 0:
sys.stderr.write(f'Warning: {self._path} invalid records (value of 9999) in global horizontal radiation\n')
number_invalid_records = self._weather_values[
self._weather_values.diffuse_horizontal_radiation_wh_m2 == 9999].count().diffuse_horizontal_radiation_wh_m2
if number_invalid_records > 0:
sys.stderr.write(f'Warning: {self._path} invalid records (value of 9999) in diffuse horizontal radiation\n')
number_invalid_records = self._weather_values[
self._weather_values.direct_normal_radiation_wh_m2 == 9999].count().direct_normal_radiation_wh_m2
if number_invalid_records > 0:
sys.stderr.write(f'Warning: {self._path} invalid records (value of 9999) in direct horizontal radiation\n')
self._weather_values = self._weather_values.to_dict(orient='list')
except SystemExit:
sys.stderr.write(f'Error: wrong formatting of weather file {self._path}\n')
sys.exit()
for building in self._city.buildings:
building.ground_temperature[cte.MONTH] = ground_temperature_from_file
ground_temperature = {}
for ground_temperature_set in building.ground_temperature[cte.MONTH]:
temperature = sum(building.ground_temperature[cte.MONTH][ground_temperature_set]) / 12
ground_temperature[ground_temperature_set] = [temperature]
building.ground_temperature[cte.YEAR] = ground_temperature
if cte.HOUR in building.external_temperature:
del building.external_temperature[cte.HOUR]
# new_value = pd.DataFrame(self._weather_values[['dry_bulb_temperature_c']].to_numpy(), columns=['epw'])
# number_invalid_records = new_value[new_value.epw == 99.9].count().epw
building.external_temperature[cte.HOUR] = self._weather_values['dry_bulb_temperature_c']
building.global_horizontal[cte.HOUR] = [x * cte.WATTS_HOUR_TO_JULES
for x in self._weather_values['global_horizontal_radiation_wh_m2']]
building.diffuse[cte.HOUR] = [x * cte.WATTS_HOUR_TO_JULES
for x in self._weather_values['diffuse_horizontal_radiation_wh_m2']]
building.direct_normal[cte.HOUR] = [x * cte.WATTS_HOUR_TO_JULES
for x in self._weather_values['direct_normal_radiation_wh_m2']]
building.beam[cte.HOUR] = [building.global_horizontal[cte.HOUR][i] -
building.diffuse[cte.HOUR][i]
for i in range(len(building.global_horizontal[cte.HOUR]))]
building.cold_water_temperature[cte.HOUR] = wh().cold_water_temperature(building.external_temperature[cte.HOUR])
# create the monthly and yearly values out of the hourly
for building in self._city.buildings:
building.external_temperature[cte.MONTH] = \
MonthlyValues().get_mean_values(building.external_temperature[cte.HOUR])
building.external_temperature[cte.YEAR] = [sum(building.external_temperature[cte.HOUR]) / 9870]
building.cold_water_temperature[cte.MONTH] = \
MonthlyValues().get_mean_values(building.cold_water_temperature[cte.HOUR])
building.cold_water_temperature[cte.YEAR] = [sum(building.cold_water_temperature[cte.HOUR]) / 9870]
# If the usage has already being imported, the domestic hot water missing values must be calculated here that
# the cold water temperature is finally known
cold_temperature = building.cold_water_temperature[cte.YEAR][0]
for internal_zone in building.internal_zones:
if internal_zone.usages is not None:
for usage in internal_zone.usages:
if usage.domestic_hot_water.peak_flow is None:
if usage.domestic_hot_water.density is None:
continue
peak_flow = 0
if (usage.domestic_hot_water.service_temperature - cold_temperature) > 0:
peak_flow = usage.domestic_hot_water.density / cte.WATER_DENSITY / cte.WATER_HEAT_CAPACITY \
/ (usage.domestic_hot_water.service_temperature - cold_temperature)
usage.domestic_hot_water.peak_flow = peak_flow
if usage.domestic_hot_water.density is None:
if usage.domestic_hot_water.peak_flow is None:
continue
density = (
usage.domestic_hot_water.peak_flow * cte.WATER_DENSITY * cte.WATER_HEAT_CAPACITY *
(usage.domestic_hot_water.service_temperature - cold_temperature)
)
usage.domestic_hot_water.density = density
self._city.level_of_detail.weather = 2
for building in self._city.buildings:
building.level_of_detail.weather = 2