Total Complexity | 121 |
Total Lines | 2830 |
Duplicated Lines | 1.55 % |
Changes | 0 |
Duplicate code is one of the most pungent code smells. A rule that is often used is to re-structure code once it is duplicated in three or more places.
Common duplication problems, and corresponding solutions are:
Complex classes like data.datasets.power_plants.pv_rooftop_buildings often do a lot of different things. To break such a class down, we need to identify a cohesive component within that class. A common approach to find such a component is to look for fields/methods that share the same prefixes, or suffixes.
Once you have determined the fields that belong together, you can apply the Extract Class refactoring. If the component makes sense as a sub-class, Extract Subclass is also a candidate, and is often faster.
1 | """ |
||
2 | Distribute MaStR PV rooftop capacities to OSM and synthetic buildings. Generate |
||
3 | new PV rooftop generators for scenarios eGon2035 and eGon100RE. |
||
4 | |||
5 | Data cleaning and inference: |
||
6 | * Drop duplicates and entries with missing critical data. |
||
7 | * Determine most plausible capacity from multiple values given in MaStR data. |
||
8 | * Drop generators which don't have any plausible capacity data |
||
9 | (23.5MW > P > 0.1). |
||
10 | * Randomly and weighted add a start-up date if it is missing. |
||
11 | * Extract zip and municipality from 'Standort' given in MaStR data. |
||
12 | * Geocode unique zip and municipality combinations with Nominatim (1 sec |
||
13 | delay). Drop generators for which geocoding failed or which are located |
||
14 | outside the municipalities of Germany. |
||
15 | * Add some visual sanity checks for cleaned data. |
||
16 | |||
17 | Allocation of MaStR data: |
||
18 | * Allocate each generator to an existing building from OSM. |
||
19 | * Determine the quantile each generator and building is in depending on the |
||
20 | capacity of the generator and the area of the polygon of the building. |
||
21 | * Randomly distribute generators within each municipality preferably within |
||
22 | the same building area quantile as the generators are capacity wise. |
||
23 | * If not enough buildings exists within a municipality and quantile additional |
||
24 | buildings from other quantiles are chosen randomly. |
||
25 | |||
26 | Desegregation of pv rooftop scenarios: |
||
27 | * The scenario data per federal state is linearly distributed to the mv grid |
||
28 | districts according to the pv rooftop potential per mv grid district. |
||
29 | * The rooftop potential is estimated from the building area given from the OSM |
||
30 | buildings. |
||
31 | * Grid districts, which are located in several federal states, are allocated |
||
32 | PV capacity according to their respective roof potential in the individual |
||
33 | federal states. |
||
34 | * The desegregation of PV plants within a grid districts respects existing |
||
35 | plants from MaStR, which did not reach their end of life. |
||
36 | * New PV plants are randomly and weighted generated using a breakdown of MaStR |
||
37 | data as generator basis. |
||
38 | * Plant metadata (e.g. plant orientation) is also added random and weighted |
||
39 | from MaStR data as basis. |
||
40 | """ |
||
41 | from __future__ import annotations |
||
42 | |||
43 | from collections import Counter |
||
44 | from functools import wraps |
||
45 | from pathlib import Path |
||
46 | from time import perf_counter |
||
47 | from typing import Any |
||
48 | |||
49 | from geoalchemy2 import Geometry |
||
50 | from geopy.extra.rate_limiter import RateLimiter |
||
51 | from geopy.geocoders import Nominatim |
||
52 | from loguru import logger |
||
53 | from numpy.random import RandomState, default_rng |
||
54 | from pyproj.crs.crs import CRS |
||
55 | from sqlalchemy import BigInteger, Column, Float, Integer, String |
||
56 | from sqlalchemy.dialects.postgresql import HSTORE |
||
57 | from sqlalchemy.ext.declarative import declarative_base |
||
58 | import geopandas as gpd |
||
59 | import numpy as np |
||
60 | import pandas as pd |
||
61 | |||
62 | from egon.data import config, db |
||
63 | from egon.data.datasets.electricity_demand_timeseries.hh_buildings import ( |
||
64 | OsmBuildingsSynthetic, |
||
65 | ) |
||
66 | from egon.data.datasets.mastr import WORKING_DIR_MASTR_NEW |
||
67 | from egon.data.datasets.power_plants.mastr import EgonPowerPlantsPv |
||
68 | from egon.data.datasets.scenario_capacities import EgonScenarioCapacities |
||
69 | from egon.data.datasets.zensus_vg250 import Vg250Gem |
||
70 | |||
71 | engine = db.engine() |
||
72 | Base = declarative_base() |
||
73 | SEED = int(config.settings()["egon-data"]["--random-seed"]) |
||
74 | |||
75 | # TODO: move to yml |
||
76 | # mastr data |
||
77 | MASTR_RELEVANT_COLS = [ |
||
78 | "EinheitMastrNummer", |
||
79 | "Bruttoleistung", |
||
80 | "Bruttoleistung_extended", |
||
81 | "Nettonennleistung", |
||
82 | "InstallierteLeistung", |
||
83 | "zugeordneteWirkleistungWechselrichter", |
||
84 | "EinheitBetriebsstatus", |
||
85 | "Standort", |
||
86 | "Bundesland", |
||
87 | "Land", |
||
88 | "Landkreis", |
||
89 | "Gemeinde", |
||
90 | "Postleitzahl", |
||
91 | "Ort", |
||
92 | "GeplantesInbetriebnahmedatum", |
||
93 | "Inbetriebnahmedatum", |
||
94 | "GemeinsamerWechselrichterMitSpeicher", |
||
95 | "Lage", |
||
96 | "Leistungsbegrenzung", |
||
97 | "EinheitlicheAusrichtungUndNeigungswinkel", |
||
98 | "Hauptausrichtung", |
||
99 | "HauptausrichtungNeigungswinkel", |
||
100 | "Nebenausrichtung", |
||
101 | ] |
||
102 | |||
103 | MASTR_DTYPES = { |
||
104 | "EinheitMastrNummer": str, |
||
105 | "Bruttoleistung": float, |
||
106 | "Bruttoleistung_extended": float, |
||
107 | "Nettonennleistung": float, |
||
108 | "InstallierteLeistung": float, |
||
109 | "zugeordneteWirkleistungWechselrichter": float, |
||
110 | "EinheitBetriebsstatus": str, |
||
111 | "Standort": str, |
||
112 | "Bundesland": str, |
||
113 | "Land": str, |
||
114 | "Landkreis": str, |
||
115 | "Gemeinde": str, |
||
116 | # "Postleitzahl": int, # fails because of nan values |
||
117 | "Ort": str, |
||
118 | "GemeinsamerWechselrichterMitSpeicher": str, |
||
119 | "Lage": str, |
||
120 | "Leistungsbegrenzung": str, |
||
121 | # this will parse nan values as false wich is not always correct |
||
122 | # "EinheitlicheAusrichtungUndNeigungswinkel": bool, |
||
123 | "Hauptausrichtung": str, |
||
124 | "HauptausrichtungNeigungswinkel": str, |
||
125 | "Nebenausrichtung": str, |
||
126 | "NebenausrichtungNeigungswinkel": str, |
||
127 | } |
||
128 | |||
129 | MASTR_PARSE_DATES = [ |
||
130 | "GeplantesInbetriebnahmedatum", |
||
131 | "Inbetriebnahmedatum", |
||
132 | ] |
||
133 | |||
134 | MASTR_INDEX_COL = "EinheitMastrNummer" |
||
135 | |||
136 | EPSG = 4326 |
||
137 | SRID = 3035 |
||
138 | |||
139 | # data cleaning |
||
140 | MAX_REALISTIC_PV_CAP = 23500 |
||
141 | MIN_REALISTIC_PV_CAP = 0.1 |
||
142 | ROUNDING = 1 |
||
143 | |||
144 | # geopy |
||
145 | MIN_DELAY_SECONDS = 1 |
||
146 | USER_AGENT = "rli_kh_geocoder" |
||
147 | |||
148 | # show additional logging information |
||
149 | VERBOSE = False |
||
150 | |||
151 | EXPORT_DIR = Path(__name__).resolve().parent / "data" |
||
152 | EXPORT_FILE = "mastr_geocoded.gpkg" |
||
153 | EXPORT_PATH = EXPORT_DIR / EXPORT_FILE |
||
154 | DRIVER = "GPKG" |
||
155 | |||
156 | # Number of quantiles |
||
157 | Q = 5 |
||
158 | |||
159 | # Scenario Data |
||
160 | CARRIER = "solar_rooftop" |
||
161 | SCENARIOS = ["eGon2035", "eGon100RE"] |
||
162 | SCENARIO_TIMESTAMP = { |
||
163 | "eGon2035": pd.Timestamp("2035-01-01", tz="UTC"), |
||
164 | "eGon100RE": pd.Timestamp("2050-01-01", tz="UTC"), |
||
165 | } |
||
166 | PV_ROOFTOP_LIFETIME = pd.Timedelta(20 * 365, unit="D") |
||
167 | |||
168 | # Example Modul Trina Vertex S TSM-400DE09M.08 400 Wp |
||
169 | # https://www.photovoltaik4all.de/media/pdf/92/64/68/Trina_Datasheet_VertexS_DE09-08_2021_A.pdf |
||
170 | MODUL_CAP = 0.4 # kWp |
||
171 | MODUL_SIZE = 1.096 * 1.754 # m² |
||
172 | PV_CAP_PER_SQ_M = MODUL_CAP / MODUL_SIZE |
||
173 | |||
174 | # Estimation of usable roof area |
||
175 | # Factor for the conversion of building area to roof area |
||
176 | # estimation mean roof pitch: 35° |
||
177 | # estimation usable roof share: 80% |
||
178 | # estimation that only the south side of the building is used for pv |
||
179 | # see https://mediatum.ub.tum.de/doc/%20969497/969497.pdf |
||
180 | # AREA_FACTOR = 1.221 |
||
181 | # USABLE_ROOF_SHARE = 0.8 |
||
182 | # SOUTH_SHARE = 0.5 |
||
183 | # ROOF_FACTOR = AREA_FACTOR * USABLE_ROOF_SHARE * SOUTH_SHARE |
||
184 | ROOF_FACTOR = 0.5 |
||
185 | |||
186 | CAP_RANGES = [ |
||
187 | (0, 30), |
||
188 | (30, 100), |
||
189 | (100, float("inf")), |
||
190 | ] |
||
191 | |||
192 | MIN_BUILDING_SIZE = 10.0 |
||
193 | UPPER_QUNATILE = 0.95 |
||
194 | LOWER_QUANTILE = 0.05 |
||
195 | |||
196 | COLS_TO_RENAME = { |
||
197 | "EinheitlicheAusrichtungUndNeigungswinkel": ( |
||
198 | "einheitliche_ausrichtung_und_neigungswinkel" |
||
199 | ), |
||
200 | "Hauptausrichtung": "hauptausrichtung", |
||
201 | "HauptausrichtungNeigungswinkel": "hauptausrichtung_neigungswinkel", |
||
202 | } |
||
203 | |||
204 | COLS_TO_EXPORT = [ |
||
205 | "scenario", |
||
206 | "bus_id", |
||
207 | "building_id", |
||
208 | "gens_id", |
||
209 | "capacity", |
||
210 | "einheitliche_ausrichtung_und_neigungswinkel", |
||
211 | "hauptausrichtung", |
||
212 | "hauptausrichtung_neigungswinkel", |
||
213 | "voltage_level", |
||
214 | "weather_cell_id", |
||
215 | ] |
||
216 | |||
217 | # TODO |
||
218 | INCLUDE_SYNTHETIC_BUILDINGS = True |
||
219 | ONLY_BUILDINGS_WITH_DEMAND = True |
||
220 | TEST_RUN = False |
||
221 | |||
222 | |||
223 | def timer_func(func): |
||
224 | @wraps(func) |
||
225 | def timeit_wrapper(*args, **kwargs): |
||
226 | start_time = perf_counter() |
||
227 | result = func(*args, **kwargs) |
||
228 | end_time = perf_counter() |
||
229 | total_time = end_time - start_time |
||
230 | logger.debug( |
||
231 | f"Function {func.__name__} took {total_time:.4f} seconds." |
||
232 | ) |
||
233 | return result |
||
234 | |||
235 | return timeit_wrapper |
||
236 | |||
237 | |||
238 | @timer_func |
||
239 | def mastr_data( |
||
240 | index_col: str | int | list[str] | list[int], |
||
241 | usecols: list[str], |
||
242 | dtype: dict[str, Any] | None, |
||
243 | parse_dates: list[str] | None, |
||
244 | ) -> pd.DataFrame: |
||
245 | """ |
||
246 | Read MaStR data from csv. |
||
247 | |||
248 | Parameters |
||
249 | ----------- |
||
250 | index_col : str, int or list of str or int |
||
251 | Column(s) to use as the row labels of the DataFrame. |
||
252 | usecols : list of str |
||
253 | Return a subset of the columns. |
||
254 | dtype : dict of column (str) -> type (any), optional |
||
255 | Data type for data or columns. |
||
256 | parse_dates : list of names (str), optional |
||
257 | Try to parse given columns to datetime. |
||
258 | Returns |
||
259 | ------- |
||
260 | pandas.DataFrame |
||
261 | DataFrame containing MaStR data. |
||
262 | """ |
||
263 | mastr_path = Path( |
||
264 | WORKING_DIR_MASTR_NEW |
||
265 | / config.datasets()["power_plants"]["sources"]["mastr_pv"] |
||
266 | ).resolve() |
||
267 | |||
268 | mastr_df = pd.read_csv( |
||
269 | mastr_path, |
||
270 | index_col=index_col, |
||
271 | usecols=usecols, |
||
272 | dtype=dtype, |
||
273 | parse_dates=parse_dates, |
||
274 | ) |
||
275 | |||
276 | mastr_df = mastr_df.loc[ |
||
277 | (mastr_df.EinheitBetriebsstatus == "InBetrieb") |
||
278 | & (mastr_df.Land == "Deutschland") |
||
279 | & (mastr_df.Lage == "Bauliche Anlagen (Hausdach, Gebäude und Fassade)") |
||
280 | ] |
||
281 | |||
282 | if ( |
||
283 | config.settings()["egon-data"]["--dataset-boundary"] |
||
284 | == "Schleswig-Holstein" |
||
285 | ): |
||
286 | init_len = len(mastr_df) |
||
287 | |||
288 | mastr_df = mastr_df.loc[mastr_df.Bundesland == "SchleswigHolstein"] |
||
289 | |||
290 | logger.info( |
||
291 | f"Using only MaStR data within Schleswig-Holstein. " |
||
292 | f"{init_len - len(mastr_df)} of {init_len} generators are dropped." |
||
293 | ) |
||
294 | |||
295 | logger.debug("MaStR data loaded.") |
||
296 | |||
297 | return mastr_df |
||
298 | |||
299 | |||
300 | @timer_func |
||
301 | def clean_mastr_data( |
||
302 | mastr_df: pd.DataFrame, |
||
303 | max_realistic_pv_cap: int | float, |
||
304 | min_realistic_pv_cap: int | float, |
||
305 | rounding: int, |
||
306 | seed: int, |
||
307 | ) -> pd.DataFrame: |
||
308 | """ |
||
309 | Clean the MaStR data from implausible data. |
||
310 | |||
311 | * Drop MaStR ID duplicates. |
||
312 | * Drop generators with implausible capacities. |
||
313 | * Drop generators without any kind of start-up date. |
||
314 | * Clean up Standort column and capacity. |
||
315 | |||
316 | Parameters |
||
317 | ----------- |
||
318 | mastr_df : pandas.DataFrame |
||
319 | DataFrame containing MaStR data. |
||
320 | max_realistic_pv_cap : int or float |
||
321 | Maximum capacity, which is considered to be realistic. |
||
322 | min_realistic_pv_cap : int or float |
||
323 | Minimum capacity, which is considered to be realistic. |
||
324 | rounding : int |
||
325 | Rounding to use when cleaning up capacity. E.g. when |
||
326 | rounding is 1 a capacity of 9.93 will be rounded to 9.9. |
||
327 | seed : int |
||
328 | Seed to use for random operations with NumPy and pandas. |
||
329 | Returns |
||
330 | ------- |
||
331 | pandas.DataFrame |
||
332 | DataFrame containing cleaned MaStR data. |
||
333 | """ |
||
334 | init_len = len(mastr_df) |
||
335 | |||
336 | # drop duplicates |
||
337 | mastr_df = mastr_df.loc[~mastr_df.index.duplicated()] |
||
338 | |||
339 | # drop invalid entries in standort |
||
340 | index_to_drop = mastr_df.loc[ |
||
341 | (mastr_df.Standort.isna()) | (mastr_df.Standort.isnull()) |
||
342 | ].index |
||
343 | |||
344 | mastr_df = mastr_df.loc[~mastr_df.index.isin(index_to_drop)] |
||
345 | |||
346 | df = mastr_df[ |
||
347 | [ |
||
348 | "Bruttoleistung", |
||
349 | "Bruttoleistung_extended", |
||
350 | "Nettonennleistung", |
||
351 | "zugeordneteWirkleistungWechselrichter", |
||
352 | "InstallierteLeistung", |
||
353 | ] |
||
354 | ].round(rounding) |
||
355 | |||
356 | # use only the smallest capacity rating if multiple are given |
||
357 | mastr_df = mastr_df.assign( |
||
358 | capacity=[ |
||
359 | most_plausible(p_tub, min_realistic_pv_cap) |
||
360 | for p_tub in df.itertuples(index=False) |
||
361 | ] |
||
362 | ) |
||
363 | |||
364 | # drop generators without any capacity info |
||
365 | # and capacity of zero |
||
366 | # and if the capacity is > 23.5 MW, because |
||
367 | # Germanies largest rooftop PV is 23 MW |
||
368 | # https://www.iwr.de/news/groesste-pv-dachanlage-europas-wird-in-sachsen-anhalt-gebaut-news37379 |
||
369 | mastr_df = mastr_df.loc[ |
||
370 | (~mastr_df.capacity.isna()) |
||
371 | & (mastr_df.capacity <= max_realistic_pv_cap) |
||
372 | & (mastr_df.capacity > min_realistic_pv_cap) |
||
373 | ] |
||
374 | |||
375 | # get zip and municipality |
||
376 | mastr_df[["zip_and_municipality", "drop_this"]] = pd.DataFrame( |
||
377 | mastr_df.Standort.astype(str) |
||
378 | .apply( |
||
379 | zip_and_municipality_from_standort, |
||
380 | args=(VERBOSE,), |
||
381 | ) |
||
382 | .tolist(), |
||
383 | index=mastr_df.index, |
||
384 | ) |
||
385 | |||
386 | # drop invalid entries |
||
387 | mastr_df = mastr_df.loc[mastr_df.drop_this].drop(columns="drop_this") |
||
388 | |||
389 | # add ", Deutschland" just in case |
||
390 | mastr_df = mastr_df.assign( |
||
391 | zip_and_municipality=(mastr_df.zip_and_municipality + ", Deutschland") |
||
392 | ) |
||
393 | |||
394 | # get consistent start-up date |
||
395 | mastr_df = mastr_df.assign( |
||
396 | start_up_date=mastr_df.Inbetriebnahmedatum, |
||
397 | ) |
||
398 | |||
399 | mastr_df.loc[mastr_df.start_up_date.isna()] = mastr_df.loc[ |
||
400 | mastr_df.start_up_date.isna() |
||
401 | ].assign( |
||
402 | start_up_date=mastr_df.GeplantesInbetriebnahmedatum.loc[ |
||
403 | mastr_df.start_up_date.isna() |
||
404 | ] |
||
405 | ) |
||
406 | |||
407 | # randomly and weighted fill missing start-up dates |
||
408 | pool = mastr_df.loc[ |
||
409 | ~mastr_df.start_up_date.isna() |
||
410 | ].start_up_date.to_numpy() |
||
411 | |||
412 | size = len(mastr_df) - len(pool) |
||
413 | |||
414 | if size > 0: |
||
415 | np.random.seed(seed) |
||
416 | |||
417 | choice = np.random.choice( |
||
418 | pool, |
||
419 | size=size, |
||
420 | replace=False, |
||
421 | ) |
||
422 | |||
423 | mastr_df.loc[mastr_df.start_up_date.isna()] = mastr_df.loc[ |
||
424 | mastr_df.start_up_date.isna() |
||
425 | ].assign(start_up_date=choice) |
||
426 | |||
427 | logger.info( |
||
428 | f"Randomly and weigthed added start-up date to {size} generators." |
||
429 | ) |
||
430 | |||
431 | mastr_df = mastr_df.assign( |
||
432 | start_up_date=pd.to_datetime(mastr_df.start_up_date, utc=True) |
||
433 | ) |
||
434 | |||
435 | end_len = len(mastr_df) |
||
436 | logger.debug( |
||
437 | f"Dropped {init_len - end_len} " |
||
438 | f"({((init_len - end_len) / init_len) * 100:g}%)" |
||
439 | f" of {init_len} rows from MaStR DataFrame." |
||
440 | ) |
||
441 | |||
442 | return mastr_df |
||
443 | |||
444 | |||
445 | def zip_and_municipality_from_standort( |
||
446 | standort: str, |
||
447 | verbose: bool = False, |
||
448 | ) -> tuple[str, bool]: |
||
449 | """ |
||
450 | Get zip code and municipality from Standort string split into a list. |
||
451 | Parameters |
||
452 | ----------- |
||
453 | standort : str |
||
454 | Standort as given from MaStR data. |
||
455 | verbose : bool |
||
456 | Logs additional info if True. |
||
457 | Returns |
||
458 | ------- |
||
459 | str |
||
460 | Standort with only the zip code and municipality |
||
461 | as well a ', Germany' added. |
||
462 | """ |
||
463 | if verbose: |
||
464 | logger.debug(f"Uncleaned String: {standort}") |
||
465 | |||
466 | standort_list = standort.split() |
||
467 | |||
468 | found = False |
||
469 | count = 0 |
||
470 | |||
471 | for count, elem in enumerate(standort_list): |
||
472 | if len(elem) != 5: |
||
473 | continue |
||
474 | if not elem.isnumeric(): |
||
475 | continue |
||
476 | |||
477 | found = True |
||
478 | |||
479 | break |
||
480 | |||
481 | if found: |
||
482 | cleaned_str = " ".join(standort_list[count:]) |
||
483 | |||
484 | if verbose: |
||
485 | logger.debug(f"Cleaned String: {cleaned_str}") |
||
486 | |||
487 | return cleaned_str, found |
||
488 | |||
489 | logger.warning( |
||
490 | "Couldn't identify zip code. This entry will be dropped." |
||
491 | f" Original standort: {standort}." |
||
492 | ) |
||
493 | |||
494 | return standort, found |
||
495 | |||
496 | |||
497 | def most_plausible( |
||
498 | p_tub: tuple, |
||
499 | min_realistic_pv_cap: int | float, |
||
500 | ) -> float: |
||
501 | """ |
||
502 | Try to determine the most plausible capacity. |
||
503 | Try to determine the most plausible capacity from a given |
||
504 | generator from MaStR data. |
||
505 | Parameters |
||
506 | ----------- |
||
507 | p_tub : tuple |
||
508 | Tuple containing the different capacities given in |
||
509 | the MaStR data. |
||
510 | min_realistic_pv_cap : int or float |
||
511 | Minimum capacity, which is considered to be realistic. |
||
512 | Returns |
||
513 | ------- |
||
514 | float |
||
515 | Capacity of the generator estimated as the most realistic. |
||
516 | """ |
||
517 | count = Counter(p_tub).most_common(3) |
||
518 | |||
519 | if len(count) == 1: |
||
520 | return count[0][0] |
||
521 | |||
522 | val1 = count[0][0] |
||
523 | val2 = count[1][0] |
||
524 | |||
525 | if len(count) == 2: |
||
526 | min_val = min(val1, val2) |
||
527 | max_val = max(val1, val2) |
||
528 | else: |
||
529 | val3 = count[2][0] |
||
530 | |||
531 | min_val = min(val1, val2, val3) |
||
532 | max_val = max(val1, val2, val3) |
||
533 | |||
534 | if min_val < min_realistic_pv_cap: |
||
535 | return max_val |
||
536 | |||
537 | return min_val |
||
538 | |||
539 | |||
540 | def geocoder( |
||
541 | user_agent: str, |
||
542 | min_delay_seconds: int, |
||
543 | ) -> RateLimiter: |
||
544 | """ |
||
545 | Setup Nominatim geocoding class. |
||
546 | Parameters |
||
547 | ----------- |
||
548 | user_agent : str |
||
549 | The app name. |
||
550 | min_delay_seconds : int |
||
551 | Delay in seconds to use between requests to Nominatim. |
||
552 | A minimum of 1 is advised. |
||
553 | Returns |
||
554 | ------- |
||
555 | geopy.extra.rate_limiter.RateLimiter |
||
556 | Nominatim RateLimiter geocoding class to use for geocoding. |
||
557 | """ |
||
558 | locator = Nominatim(user_agent=user_agent) |
||
559 | return RateLimiter( |
||
560 | locator.geocode, |
||
561 | min_delay_seconds=min_delay_seconds, |
||
562 | ) |
||
563 | |||
564 | |||
565 | def geocoding_data( |
||
566 | clean_mastr_df: pd.DataFrame, |
||
567 | ) -> pd.DataFrame: |
||
568 | """ |
||
569 | Setup DataFrame to geocode. |
||
570 | Parameters |
||
571 | ----------- |
||
572 | clean_mastr_df : pandas.DataFrame |
||
573 | DataFrame containing cleaned MaStR data. |
||
574 | Returns |
||
575 | ------- |
||
576 | pandas.DataFrame |
||
577 | DataFrame containing all unique combinations of |
||
578 | zip codes with municipalities for geocoding. |
||
579 | """ |
||
580 | return pd.DataFrame( |
||
581 | data=clean_mastr_df.zip_and_municipality.unique(), |
||
582 | columns=["zip_and_municipality"], |
||
583 | ) |
||
584 | |||
585 | |||
586 | @timer_func |
||
587 | def geocode_data( |
||
588 | geocoding_df: pd.DataFrame, |
||
589 | ratelimiter: RateLimiter, |
||
590 | epsg: int, |
||
591 | ) -> gpd.GeoDataFrame: |
||
592 | """ |
||
593 | Geocode zip code and municipality. |
||
594 | Extract latitude, longitude and altitude. |
||
595 | Transfrom latitude and longitude to shapely |
||
596 | Point and return a geopandas GeoDataFrame. |
||
597 | Parameters |
||
598 | ----------- |
||
599 | geocoding_df : pandas.DataFrame |
||
600 | DataFrame containing all unique combinations of |
||
601 | zip codes with municipalities for geocoding. |
||
602 | ratelimiter : geopy.extra.rate_limiter.RateLimiter |
||
603 | Nominatim RateLimiter geocoding class to use for geocoding. |
||
604 | epsg : int |
||
605 | EPSG ID to use as CRS. |
||
606 | Returns |
||
607 | ------- |
||
608 | geopandas.GeoDataFrame |
||
609 | GeoDataFrame containing all unique combinations of |
||
610 | zip codes with municipalities with matching geolocation. |
||
611 | """ |
||
612 | logger.info(f"Geocoding {len(geocoding_df)} locations.") |
||
613 | |||
614 | geocode_df = geocoding_df.assign( |
||
615 | location=geocoding_df.zip_and_municipality.apply(ratelimiter) |
||
616 | ) |
||
617 | |||
618 | geocode_df = geocode_df.assign( |
||
619 | point=geocode_df.location.apply( |
||
620 | lambda loc: tuple(loc.point) if loc else None |
||
621 | ) |
||
622 | ) |
||
623 | |||
624 | geocode_df[["latitude", "longitude", "altitude"]] = pd.DataFrame( |
||
625 | geocode_df.point.tolist(), index=geocode_df.index |
||
626 | ) |
||
627 | |||
628 | return gpd.GeoDataFrame( |
||
629 | geocode_df, |
||
630 | geometry=gpd.points_from_xy(geocode_df.longitude, geocode_df.latitude), |
||
631 | crs=f"EPSG:{epsg}", |
||
632 | ) |
||
633 | |||
634 | |||
635 | def merge_geocode_with_mastr( |
||
636 | clean_mastr_df: pd.DataFrame, geocode_gdf: gpd.GeoDataFrame |
||
637 | ) -> gpd.GeoDataFrame: |
||
638 | """ |
||
639 | Merge geometry to original mastr data. |
||
640 | Parameters |
||
641 | ----------- |
||
642 | clean_mastr_df : pandas.DataFrame |
||
643 | DataFrame containing cleaned MaStR data. |
||
644 | geocode_gdf : geopandas.GeoDataFrame |
||
645 | GeoDataFrame containing all unique combinations of |
||
646 | zip codes with municipalities with matching geolocation. |
||
647 | Returns |
||
648 | ------- |
||
649 | gepandas.GeoDataFrame |
||
650 | GeoDataFrame containing cleaned MaStR data with |
||
651 | matching geolocation from geocoding. |
||
652 | """ |
||
653 | return gpd.GeoDataFrame( |
||
654 | clean_mastr_df.merge( |
||
655 | geocode_gdf[["zip_and_municipality", "geometry"]], |
||
656 | how="left", |
||
657 | left_on="zip_and_municipality", |
||
658 | right_on="zip_and_municipality", |
||
659 | ), |
||
660 | crs=geocode_gdf.crs, |
||
661 | ).set_index(clean_mastr_df.index) |
||
662 | |||
663 | |||
664 | def drop_invalid_entries_from_gdf( |
||
665 | gdf: gpd.GeoDataFrame, |
||
666 | ) -> gpd.GeoDataFrame: |
||
667 | """ |
||
668 | Drop invalid entries from geopandas GeoDataFrame. |
||
669 | TODO: how to omit the logging from geos here??? |
||
670 | Parameters |
||
671 | ----------- |
||
672 | gdf : geopandas.GeoDataFrame |
||
673 | GeoDataFrame to be checked for validity. |
||
674 | Returns |
||
675 | ------- |
||
676 | gepandas.GeoDataFrame |
||
677 | GeoDataFrame with rows with invalid geometries |
||
678 | dropped. |
||
679 | """ |
||
680 | valid_gdf = gdf.loc[gdf.is_valid] |
||
681 | |||
682 | logger.debug( |
||
683 | f"{len(gdf) - len(valid_gdf)} " |
||
684 | f"({(len(gdf) - len(valid_gdf)) / len(gdf) * 100:g}%) " |
||
685 | f"of {len(gdf)} values were invalid and are dropped." |
||
686 | ) |
||
687 | |||
688 | return valid_gdf |
||
689 | |||
690 | |||
691 | @timer_func |
||
692 | def municipality_data() -> gpd.GeoDataFrame: |
||
693 | """ |
||
694 | Get municipality data from eGo^n Database. |
||
695 | Returns |
||
696 | ------- |
||
697 | gepandas.GeoDataFrame |
||
698 | GeoDataFrame with municipality data. |
||
699 | """ |
||
700 | with db.session_scope() as session: |
||
701 | query = session.query(Vg250Gem.ags, Vg250Gem.geometry.label("geom")) |
||
702 | |||
703 | return gpd.read_postgis( |
||
704 | query.statement, query.session.bind, index_col="ags" |
||
705 | ) |
||
706 | |||
707 | |||
708 | @timer_func |
||
709 | def add_ags_to_gens( |
||
710 | valid_mastr_gdf: gpd.GeoDataFrame, |
||
711 | municipalities_gdf: gpd.GeoDataFrame, |
||
712 | ) -> gpd.GeoDataFrame: |
||
713 | """ |
||
714 | Add information about AGS ID to generators. |
||
715 | Parameters |
||
716 | ----------- |
||
717 | valid_mastr_gdf : geopandas.GeoDataFrame |
||
718 | GeoDataFrame with valid and cleaned MaStR data. |
||
719 | municipalities_gdf : geopandas.GeoDataFrame |
||
720 | GeoDataFrame with municipality data. |
||
721 | Returns |
||
722 | ------- |
||
723 | gepandas.GeoDataFrame |
||
724 | GeoDataFrame with valid and cleaned MaStR data |
||
725 | with AGS ID added. |
||
726 | """ |
||
727 | return valid_mastr_gdf.sjoin( |
||
728 | municipalities_gdf, |
||
729 | how="left", |
||
730 | predicate="intersects", |
||
731 | ).rename(columns={"index_right": "ags"}) |
||
732 | |||
733 | |||
734 | def drop_gens_outside_muns( |
||
735 | valid_mastr_gdf: gpd.GeoDataFrame, |
||
736 | ) -> gpd.GeoDataFrame: |
||
737 | """ |
||
738 | Drop all generators outside of municipalities. |
||
739 | Parameters |
||
740 | ----------- |
||
741 | valid_mastr_gdf : geopandas.GeoDataFrame |
||
742 | GeoDataFrame with valid and cleaned MaStR data. |
||
743 | Returns |
||
744 | ------- |
||
745 | gepandas.GeoDataFrame |
||
746 | GeoDataFrame with valid and cleaned MaStR data |
||
747 | with generatos without an AGS ID dropped. |
||
748 | """ |
||
749 | gdf = valid_mastr_gdf.loc[~valid_mastr_gdf.ags.isna()] |
||
750 | |||
751 | logger.debug( |
||
752 | f"{len(valid_mastr_gdf) - len(gdf)} " |
||
753 | f"({(len(valid_mastr_gdf) - len(gdf)) / len(valid_mastr_gdf) * 100:g}%) " |
||
754 | f"of {len(valid_mastr_gdf)} values are outside of the municipalities" |
||
755 | " and are therefore dropped." |
||
756 | ) |
||
757 | |||
758 | return gdf |
||
759 | |||
760 | |||
761 | class EgonMastrPvRoofGeocoded(Base): |
||
762 | __tablename__ = "egon_mastr_pv_roof_geocoded" |
||
763 | __table_args__ = {"schema": "supply"} |
||
764 | |||
765 | zip_and_municipality = Column(String, primary_key=True, index=True) |
||
766 | location = Column(String) |
||
767 | point = Column(String) |
||
768 | latitude = Column(Float) |
||
769 | longitude = Column(Float) |
||
770 | altitude = Column(Float) |
||
771 | geometry = Column(Geometry(srid=EPSG)) |
||
772 | |||
773 | |||
774 | def create_geocoded_table(geocode_gdf): |
||
775 | """ |
||
776 | Create geocoded table mastr pv rooftop |
||
777 | Parameters |
||
778 | ----------- |
||
779 | geocode_gdf : geopandas.GeoDataFrame |
||
780 | GeoDataFrame containing geocoding information on pv rooftop locations. |
||
781 | """ |
||
782 | EgonMastrPvRoofGeocoded.__table__.drop(bind=engine, checkfirst=True) |
||
783 | EgonMastrPvRoofGeocoded.__table__.create(bind=engine, checkfirst=True) |
||
784 | |||
785 | geocode_gdf.to_postgis( |
||
786 | name=EgonMastrPvRoofGeocoded.__table__.name, |
||
787 | schema=EgonMastrPvRoofGeocoded.__table__.schema, |
||
788 | con=db.engine(), |
||
789 | if_exists="append", |
||
790 | index=False, |
||
791 | # dtype={} |
||
792 | ) |
||
793 | |||
794 | |||
795 | def geocoded_data_from_db( |
||
796 | epsg: str | int, |
||
797 | ) -> gpd.GeoDataFrame: |
||
798 | """ |
||
799 | Read OSM buildings data from eGo^n Database. |
||
800 | Parameters |
||
801 | ----------- |
||
802 | to_crs : pyproj.crs.crs.CRS |
||
803 | CRS to transform geometries to. |
||
804 | Returns |
||
805 | ------- |
||
806 | geopandas.GeoDataFrame |
||
807 | GeoDataFrame containing OSM buildings data. |
||
808 | """ |
||
809 | with db.session_scope() as session: |
||
810 | query = session.query( |
||
811 | EgonMastrPvRoofGeocoded.zip_and_municipality, |
||
812 | EgonMastrPvRoofGeocoded.geometry, |
||
813 | ) |
||
814 | |||
815 | return gpd.read_postgis( |
||
816 | query.statement, query.session.bind, geom_col="geometry" |
||
817 | ).to_crs(f"EPSG:{epsg}") |
||
818 | |||
819 | |||
820 | def load_mastr_data(): |
||
821 | """Read PV rooftop data from MaStR CSV |
||
822 | Note: the source will be replaced as soon as the MaStR data is available |
||
823 | in DB. |
||
824 | Returns |
||
825 | ------- |
||
826 | geopandas.GeoDataFrame |
||
827 | GeoDataFrame containing MaStR data with geocoded locations. |
||
828 | """ |
||
829 | mastr_df = mastr_data( |
||
830 | MASTR_INDEX_COL, |
||
831 | MASTR_RELEVANT_COLS, |
||
832 | MASTR_DTYPES, |
||
833 | MASTR_PARSE_DATES, |
||
834 | ) |
||
835 | |||
836 | clean_mastr_df = clean_mastr_data( |
||
837 | mastr_df, |
||
838 | max_realistic_pv_cap=MAX_REALISTIC_PV_CAP, |
||
839 | min_realistic_pv_cap=MIN_REALISTIC_PV_CAP, |
||
840 | seed=SEED, |
||
841 | rounding=ROUNDING, |
||
842 | ) |
||
843 | |||
844 | geocode_gdf = geocoded_data_from_db(EPSG) |
||
845 | |||
846 | mastr_gdf = merge_geocode_with_mastr(clean_mastr_df, geocode_gdf) |
||
847 | |||
848 | valid_mastr_gdf = drop_invalid_entries_from_gdf(mastr_gdf) |
||
849 | |||
850 | municipalities_gdf = municipality_data() |
||
851 | |||
852 | valid_mastr_gdf = add_ags_to_gens(valid_mastr_gdf, municipalities_gdf) |
||
853 | |||
854 | return drop_gens_outside_muns(valid_mastr_gdf) |
||
855 | |||
856 | |||
857 | class OsmBuildingsFiltered(Base): |
||
858 | __tablename__ = "osm_buildings_filtered" |
||
859 | __table_args__ = {"schema": "openstreetmap"} |
||
860 | |||
861 | osm_id = Column(BigInteger) |
||
862 | amenity = Column(String) |
||
863 | building = Column(String) |
||
864 | name = Column(String) |
||
865 | geom = Column(Geometry(srid=SRID), index=True) |
||
866 | area = Column(Float) |
||
867 | geom_point = Column(Geometry(srid=SRID), index=True) |
||
868 | tags = Column(HSTORE) |
||
869 | id = Column(BigInteger, primary_key=True, index=True) |
||
870 | |||
871 | |||
872 | @timer_func |
||
873 | def osm_buildings( |
||
874 | to_crs: CRS, |
||
875 | ) -> gpd.GeoDataFrame: |
||
876 | """ |
||
877 | Read OSM buildings data from eGo^n Database. |
||
878 | Parameters |
||
879 | ----------- |
||
880 | to_crs : pyproj.crs.crs.CRS |
||
881 | CRS to transform geometries to. |
||
882 | Returns |
||
883 | ------- |
||
884 | geopandas.GeoDataFrame |
||
885 | GeoDataFrame containing OSM buildings data. |
||
886 | """ |
||
887 | with db.session_scope() as session: |
||
888 | query = session.query( |
||
889 | OsmBuildingsFiltered.id, |
||
890 | OsmBuildingsFiltered.area, |
||
891 | OsmBuildingsFiltered.geom_point.label("geom"), |
||
892 | ) |
||
893 | |||
894 | return gpd.read_postgis( |
||
895 | query.statement, query.session.bind, index_col="id" |
||
896 | ).to_crs(to_crs) |
||
897 | |||
898 | |||
899 | @timer_func |
||
900 | def synthetic_buildings( |
||
901 | to_crs: CRS, |
||
902 | ) -> gpd.GeoDataFrame: |
||
903 | """ |
||
904 | Read synthetic buildings data from eGo^n Database. |
||
905 | Parameters |
||
906 | ----------- |
||
907 | to_crs : pyproj.crs.crs.CRS |
||
908 | CRS to transform geometries to. |
||
909 | Returns |
||
910 | ------- |
||
911 | geopandas.GeoDataFrame |
||
912 | GeoDataFrame containing OSM buildings data. |
||
913 | """ |
||
914 | with db.session_scope() as session: |
||
915 | query = session.query( |
||
916 | OsmBuildingsSynthetic.id, |
||
917 | OsmBuildingsSynthetic.area, |
||
918 | OsmBuildingsSynthetic.geom_point.label("geom"), |
||
919 | ) |
||
920 | |||
921 | return gpd.read_postgis( |
||
922 | query.statement, query.session.bind, index_col="id" |
||
923 | ).to_crs(to_crs) |
||
924 | |||
925 | |||
926 | @timer_func |
||
927 | def add_ags_to_buildings( |
||
928 | buildings_gdf: gpd.GeoDataFrame, |
||
929 | municipalities_gdf: gpd.GeoDataFrame, |
||
930 | ) -> gpd.GeoDataFrame: |
||
931 | """ |
||
932 | Add information about AGS ID to buildings. |
||
933 | Parameters |
||
934 | ----------- |
||
935 | buildings_gdf : geopandas.GeoDataFrame |
||
936 | GeoDataFrame containing OSM buildings data. |
||
937 | municipalities_gdf : geopandas.GeoDataFrame |
||
938 | GeoDataFrame with municipality data. |
||
939 | Returns |
||
940 | ------- |
||
941 | gepandas.GeoDataFrame |
||
942 | GeoDataFrame containing OSM buildings data |
||
943 | with AGS ID added. |
||
944 | """ |
||
945 | return buildings_gdf.sjoin( |
||
946 | municipalities_gdf, |
||
947 | how="left", |
||
948 | predicate="intersects", |
||
949 | ).rename(columns={"index_right": "ags"}) |
||
950 | |||
951 | |||
952 | def drop_buildings_outside_muns( |
||
953 | buildings_gdf: gpd.GeoDataFrame, |
||
954 | ) -> gpd.GeoDataFrame: |
||
955 | """ |
||
956 | Drop all buildings outside of municipalities. |
||
957 | Parameters |
||
958 | ----------- |
||
959 | buildings_gdf : geopandas.GeoDataFrame |
||
960 | GeoDataFrame containing OSM buildings data. |
||
961 | Returns |
||
962 | ------- |
||
963 | gepandas.GeoDataFrame |
||
964 | GeoDataFrame containing OSM buildings data |
||
965 | with buildings without an AGS ID dropped. |
||
966 | """ |
||
967 | gdf = buildings_gdf.loc[~buildings_gdf.ags.isna()] |
||
968 | |||
969 | logger.debug( |
||
970 | f"{len(buildings_gdf) - len(gdf)} " |
||
971 | f"({(len(buildings_gdf) - len(gdf)) / len(buildings_gdf) * 100:g}%) " |
||
972 | f"of {len(buildings_gdf)} values are outside of the municipalities " |
||
973 | "and are therefore dropped." |
||
974 | ) |
||
975 | |||
976 | return gdf |
||
977 | |||
978 | |||
979 | def egon_building_peak_loads(): |
||
980 | sql = """ |
||
981 | SELECT building_id |
||
982 | FROM demand.egon_building_electricity_peak_loads |
||
983 | WHERE scenario = 'eGon2035' |
||
984 | """ |
||
985 | |||
986 | return ( |
||
987 | db.select_dataframe(sql).building_id.astype(int).sort_values().unique() |
||
988 | ) |
||
989 | |||
990 | |||
991 | @timer_func |
||
992 | def load_building_data(): |
||
993 | """ |
||
994 | Read buildings from DB |
||
995 | Tables: |
||
996 | |||
997 | * `openstreetmap.osm_buildings_filtered` (from OSM) |
||
998 | * `openstreetmap.osm_buildings_synthetic` (synthetic, created by us) |
||
999 | |||
1000 | Use column `id` for both as it is unique hence you concat both datasets. |
||
1001 | If INCLUDE_SYNTHETIC_BUILDINGS is False synthetic buildings will not be |
||
1002 | loaded. |
||
1003 | |||
1004 | Returns |
||
1005 | ------- |
||
1006 | gepandas.GeoDataFrame |
||
1007 | GeoDataFrame containing OSM buildings data with buildings without an |
||
1008 | AGS ID dropped. |
||
1009 | """ |
||
1010 | |||
1011 | municipalities_gdf = municipality_data() |
||
1012 | |||
1013 | osm_buildings_gdf = osm_buildings(municipalities_gdf.crs) |
||
1014 | |||
1015 | if INCLUDE_SYNTHETIC_BUILDINGS: |
||
1016 | synthetic_buildings_gdf = synthetic_buildings(municipalities_gdf.crs) |
||
1017 | |||
1018 | buildings_gdf = gpd.GeoDataFrame( |
||
1019 | pd.concat( |
||
1020 | [ |
||
1021 | osm_buildings_gdf, |
||
1022 | synthetic_buildings_gdf, |
||
1023 | ] |
||
1024 | ), |
||
1025 | geometry="geom", |
||
1026 | crs=osm_buildings_gdf.crs, |
||
1027 | ).rename(columns={"area": "building_area"}) |
||
1028 | |||
1029 | buildings_gdf.index = buildings_gdf.index.astype(int) |
||
1030 | |||
1031 | else: |
||
1032 | buildings_gdf = osm_buildings_gdf.rename( |
||
1033 | columns={"area": "building_area"} |
||
1034 | ) |
||
1035 | |||
1036 | if ONLY_BUILDINGS_WITH_DEMAND: |
||
1037 | building_ids = egon_building_peak_loads() |
||
1038 | |||
1039 | init_len = len(building_ids) |
||
1040 | |||
1041 | building_ids = np.intersect1d( |
||
1042 | list(map(int, building_ids)), |
||
1043 | list(map(int, buildings_gdf.index.to_numpy())), |
||
1044 | ) |
||
1045 | |||
1046 | end_len = len(building_ids) |
||
1047 | |||
1048 | logger.debug( |
||
1049 | f"{end_len/init_len * 100: g} % ({end_len} / {init_len}) " |
||
1050 | f"of buildings have peak load." |
||
1051 | ) |
||
1052 | |||
1053 | buildings_gdf = buildings_gdf.loc[building_ids] |
||
1054 | |||
1055 | buildings_ags_gdf = add_ags_to_buildings(buildings_gdf, municipalities_gdf) |
||
1056 | |||
1057 | buildings_ags_gdf = drop_buildings_outside_muns(buildings_ags_gdf) |
||
1058 | |||
1059 | grid_districts_gdf = grid_districts(EPSG) |
||
1060 | |||
1061 | federal_state_gdf = federal_state_data(grid_districts_gdf.crs) |
||
1062 | |||
1063 | grid_federal_state_gdf = overlay_grid_districts_with_counties( |
||
1064 | grid_districts_gdf, |
||
1065 | federal_state_gdf, |
||
1066 | ) |
||
1067 | |||
1068 | buildings_overlay_gdf = add_overlay_id_to_buildings( |
||
1069 | buildings_ags_gdf, |
||
1070 | grid_federal_state_gdf, |
||
1071 | ) |
||
1072 | |||
1073 | logger.debug("Loaded buildings.") |
||
1074 | |||
1075 | buildings_overlay_gdf = drop_buildings_outside_grids(buildings_overlay_gdf) |
||
1076 | |||
1077 | # overwrite bus_id with data from new table |
||
1078 | sql = ( |
||
1079 | "SELECT building_id, bus_id FROM " |
||
1080 | "boundaries.egon_map_zensus_mvgd_buildings" |
||
1081 | ) |
||
1082 | map_building_bus_df = db.select_dataframe(sql) |
||
1083 | |||
1084 | building_ids = np.intersect1d( |
||
1085 | list(map(int, map_building_bus_df.building_id.unique())), |
||
1086 | list(map(int, buildings_overlay_gdf.index.to_numpy())), |
||
1087 | ) |
||
1088 | |||
1089 | buildings_within_gdf = buildings_overlay_gdf.loc[building_ids] |
||
1090 | |||
1091 | gdf = ( |
||
1092 | buildings_within_gdf.reset_index() |
||
1093 | .drop(columns=["bus_id"]) |
||
1094 | .merge( |
||
1095 | how="left", |
||
1096 | right=map_building_bus_df, |
||
1097 | left_on="id", |
||
1098 | right_on="building_id", |
||
1099 | ) |
||
1100 | .drop(columns=["building_id"]) |
||
1101 | .set_index("id") |
||
1102 | .sort_index() |
||
1103 | ) |
||
1104 | |||
1105 | return gdf[~gdf.index.duplicated(keep="first")] |
||
1106 | |||
1107 | |||
1108 | @timer_func |
||
1109 | def sort_and_qcut_df( |
||
1110 | df: pd.DataFrame | gpd.GeoDataFrame, |
||
1111 | col: str, |
||
1112 | q: int, |
||
1113 | ) -> pd.DataFrame | gpd.GeoDataFrame: |
||
1114 | """ |
||
1115 | Determine the quantile of a given attribute in a (Geo)DataFrame. |
||
1116 | Sort the (Geo)DataFrame in ascending order for the given attribute. |
||
1117 | Parameters |
||
1118 | ----------- |
||
1119 | df : pandas.DataFrame or geopandas.GeoDataFrame |
||
1120 | (Geo)DataFrame to sort and qcut. |
||
1121 | col : str |
||
1122 | Name of the attribute to sort and qcut the (Geo)DataFrame on. |
||
1123 | q : int |
||
1124 | Number of quantiles. |
||
1125 | Returns |
||
1126 | ------- |
||
1127 | pandas.DataFrame or gepandas.GeoDataFrame |
||
1128 | Sorted and qcut (Geo)DataFrame. |
||
1129 | """ |
||
1130 | df = df.sort_values(col, ascending=True) |
||
1131 | |||
1132 | return df.assign( |
||
1133 | quant=pd.qcut( |
||
1134 | df[col], |
||
1135 | q=q, |
||
1136 | labels=range(q), |
||
1137 | ) |
||
1138 | ) |
||
1139 | |||
1140 | |||
1141 | @timer_func |
||
1142 | def allocate_pv( |
||
1143 | q_mastr_gdf: gpd.GeoDataFrame, |
||
1144 | q_buildings_gdf: gpd.GeoDataFrame, |
||
1145 | seed: int, |
||
1146 | ) -> tuple[gpd.GeoDataFrame, gpd.GeoDataFrame]: |
||
1147 | """ |
||
1148 | Allocate the MaStR pv generators to the OSM buildings. |
||
1149 | This will determine a building for each pv generator if there are more |
||
1150 | buildings than generators within a given AGS. Primarily generators are |
||
1151 | distributed with the same qunatile as the buildings. Multiple assignment |
||
1152 | is excluded. |
||
1153 | Parameters |
||
1154 | ----------- |
||
1155 | q_mastr_gdf : geopandas.GeoDataFrame |
||
1156 | GeoDataFrame containing geocoded and qcut MaStR data. |
||
1157 | q_buildings_gdf : geopandas.GeoDataFrame |
||
1158 | GeoDataFrame containing qcut OSM buildings data. |
||
1159 | seed : int |
||
1160 | Seed to use for random operations with NumPy and pandas. |
||
1161 | Returns |
||
1162 | ------- |
||
1163 | tuple with two geopandas.GeoDataFrame s |
||
1164 | GeoDataFrame containing MaStR data allocated to building IDs. |
||
1165 | GeoDataFrame containing building data allocated to MaStR IDs. |
||
1166 | """ |
||
1167 | rng = default_rng(seed=seed) |
||
1168 | |||
1169 | q_buildings_gdf = q_buildings_gdf.assign(gens_id=np.nan).sort_values( |
||
1170 | by=["ags", "quant"] |
||
1171 | ) |
||
1172 | q_mastr_gdf = q_mastr_gdf.assign(building_id=np.nan).sort_values( |
||
1173 | by=["ags", "quant"] |
||
1174 | ) |
||
1175 | |||
1176 | ags_list = q_buildings_gdf.ags.unique() |
||
1177 | |||
1178 | if TEST_RUN: |
||
1179 | ags_list = ags_list[:250] |
||
1180 | |||
1181 | num_ags = len(ags_list) |
||
1182 | |||
1183 | t0 = perf_counter() |
||
1184 | |||
1185 | for count, ags in enumerate(ags_list): |
||
1186 | |||
1187 | buildings = q_buildings_gdf.loc[q_buildings_gdf.ags == ags] |
||
1188 | gens = q_mastr_gdf.loc[q_mastr_gdf.ags == ags] |
||
1189 | |||
1190 | len_build = len(buildings) |
||
1191 | len_gens = len(gens) |
||
1192 | |||
1193 | if len_build < len_gens: |
||
1194 | gens = gens.sample(len_build, random_state=RandomState(seed=seed)) |
||
1195 | logger.error( |
||
1196 | f"There are {len_gens} generators and only {len_build}" |
||
1197 | f" buildings in AGS {ags}. {len_gens - len(gens)} " |
||
1198 | "generators were truncated to match the amount of buildings." |
||
1199 | ) |
||
1200 | |||
1201 | assert len_build == len(gens) |
||
1202 | |||
1203 | for quant in gens.quant.unique(): |
||
1204 | q_buildings = buildings.loc[buildings.quant == quant] |
||
1205 | q_gens = gens.loc[gens.quant == quant] |
||
1206 | |||
1207 | len_build = len(q_buildings) |
||
1208 | len_gens = len(q_gens) |
||
1209 | |||
1210 | if len_build < len_gens: |
||
1211 | delta = len_gens - len_build |
||
1212 | |||
1213 | logger.warning( |
||
1214 | f"There are {len_gens} generators and only {len_build} " |
||
1215 | f"buildings in AGS {ags} and quantile {quant}. {delta} " |
||
1216 | f"buildings from AGS {ags} will be added randomly." |
||
1217 | ) |
||
1218 | |||
1219 | add_buildings = pd.Index( |
||
1220 | rng.choice( |
||
1221 | list(set(buildings.index) - set(q_buildings.index)), |
||
1222 | size=delta, |
||
1223 | replace=False, |
||
1224 | ) |
||
1225 | ) |
||
1226 | |||
1227 | chosen_buildings = q_buildings.index.append(add_buildings) |
||
1228 | |||
1229 | else: |
||
1230 | chosen_buildings = rng.choice( |
||
1231 | q_buildings.index, |
||
1232 | size=len_gens, |
||
1233 | replace=False, |
||
1234 | ) |
||
1235 | |||
1236 | q_buildings_gdf.loc[chosen_buildings, "gens_id"] = q_gens.index |
||
1237 | buildings = buildings.drop(chosen_buildings) |
||
1238 | |||
1239 | if count % 500 == 0: |
||
1240 | logger.debug( |
||
1241 | f"Allocation of {count / num_ags * 100:g} % of AGS done. " |
||
1242 | f"It took {perf_counter() - t0:g} seconds." |
||
1243 | ) |
||
1244 | |||
1245 | t0 = perf_counter() |
||
1246 | |||
1247 | assigned_buildings = q_buildings_gdf.loc[~q_buildings_gdf.gens_id.isna()] |
||
1248 | |||
1249 | assert len(assigned_buildings) == len(assigned_buildings.gens_id.unique()) |
||
1250 | |||
1251 | q_mastr_gdf.loc[ |
||
1252 | assigned_buildings.gens_id, "building_id" |
||
1253 | ] = assigned_buildings.index |
||
1254 | |||
1255 | assigned_gens = q_mastr_gdf.loc[~q_mastr_gdf.building_id.isna()] |
||
1256 | |||
1257 | assert len(assigned_buildings) == len(assigned_gens) |
||
1258 | |||
1259 | logger.debug("Allocated status quo generators to buildings.") |
||
1260 | |||
1261 | return frame_to_numeric(q_mastr_gdf), frame_to_numeric(q_buildings_gdf) |
||
1262 | |||
1263 | |||
1264 | def frame_to_numeric( |
||
1265 | df: pd.DataFrame | gpd.GeoDataFrame, |
||
1266 | ) -> pd.DataFrame | gpd.GeoDataFrame: |
||
1267 | """ |
||
1268 | Try to convert all columns of a DataFrame to numeric ignoring errors. |
||
1269 | Parameters |
||
1270 | ---------- |
||
1271 | df : pandas.DataFrame or geopandas.GeoDataFrame |
||
1272 | Returns |
||
1273 | ------- |
||
1274 | pandas.DataFrame or geopandas.GeoDataFrame |
||
1275 | """ |
||
1276 | if str(df.index.dtype) == "object": |
||
1277 | df.index = pd.to_numeric(df.index, errors="ignore") |
||
1278 | |||
1279 | for col in df.columns: |
||
1280 | if str(df[col].dtype) == "object": |
||
1281 | df[col] = pd.to_numeric(df[col], errors="ignore") |
||
1282 | |||
1283 | return df |
||
1284 | |||
1285 | |||
1286 | def validate_output( |
||
1287 | desagg_mastr_gdf: pd.DataFrame | gpd.GeoDataFrame, |
||
1288 | desagg_buildings_gdf: pd.DataFrame | gpd.GeoDataFrame, |
||
1289 | ) -> None: |
||
1290 | """ |
||
1291 | Validate output. |
||
1292 | |||
1293 | * Validate that there are exactly as many buildings with a pv system as |
||
1294 | there are pv systems with a building |
||
1295 | * Validate that the building IDs with a pv system are the same building |
||
1296 | IDs as assigned to the pv systems |
||
1297 | * Validate that the pv system IDs with a building are the same pv system |
||
1298 | IDs as assigned to the buildings |
||
1299 | |||
1300 | Parameters |
||
1301 | ----------- |
||
1302 | desagg_mastr_gdf : geopandas.GeoDataFrame |
||
1303 | GeoDataFrame containing MaStR data allocated to building IDs. |
||
1304 | desagg_buildings_gdf : geopandas.GeoDataFrame |
||
1305 | GeoDataFrame containing building data allocated to MaStR IDs. |
||
1306 | """ |
||
1307 | assert len( |
||
1308 | desagg_mastr_gdf.loc[~desagg_mastr_gdf.building_id.isna()] |
||
1309 | ) == len(desagg_buildings_gdf.loc[~desagg_buildings_gdf.gens_id.isna()]) |
||
1310 | assert ( |
||
1311 | np.sort( |
||
1312 | desagg_mastr_gdf.loc[ |
||
1313 | ~desagg_mastr_gdf.building_id.isna() |
||
1314 | ].building_id.unique() |
||
1315 | ) |
||
1316 | == np.sort( |
||
1317 | desagg_buildings_gdf.loc[ |
||
1318 | ~desagg_buildings_gdf.gens_id.isna() |
||
1319 | ].index.unique() |
||
1320 | ) |
||
1321 | ).all() |
||
1322 | assert ( |
||
1323 | np.sort( |
||
1324 | desagg_mastr_gdf.loc[ |
||
1325 | ~desagg_mastr_gdf.building_id.isna() |
||
1326 | ].index.unique() |
||
1327 | ) |
||
1328 | == np.sort( |
||
1329 | desagg_buildings_gdf.loc[ |
||
1330 | ~desagg_buildings_gdf.gens_id.isna() |
||
1331 | ].gens_id.unique() |
||
1332 | ) |
||
1333 | ).all() |
||
1334 | |||
1335 | logger.debug("Validated output.") |
||
1336 | |||
1337 | |||
1338 | def drop_unallocated_gens( |
||
1339 | gdf: gpd.GeoDataFrame, |
||
1340 | ) -> gpd.GeoDataFrame: |
||
1341 | """ |
||
1342 | Drop generators which did not get allocated. |
||
1343 | |||
1344 | Parameters |
||
1345 | ----------- |
||
1346 | gdf : geopandas.GeoDataFrame |
||
1347 | GeoDataFrame containing MaStR data allocated to building IDs. |
||
1348 | Returns |
||
1349 | ------- |
||
1350 | geopandas.GeoDataFrame |
||
1351 | GeoDataFrame containing MaStR data with generators dropped which did |
||
1352 | not get allocated. |
||
1353 | """ |
||
1354 | init_len = len(gdf) |
||
1355 | gdf = gdf.loc[~gdf.building_id.isna()] |
||
1356 | end_len = len(gdf) |
||
1357 | |||
1358 | logger.debug( |
||
1359 | f"Dropped {init_len - end_len} " |
||
1360 | f"({((init_len - end_len) / init_len) * 100:g}%)" |
||
1361 | f" of {init_len} unallocated rows from MaStR DataFrame." |
||
1362 | ) |
||
1363 | |||
1364 | return gdf |
||
1365 | |||
1366 | |||
1367 | @timer_func |
||
1368 | def allocate_to_buildings( |
||
1369 | mastr_gdf: gpd.GeoDataFrame, |
||
1370 | buildings_gdf: gpd.GeoDataFrame, |
||
1371 | ) -> tuple[gpd.GeoDataFrame, gpd.GeoDataFrame]: |
||
1372 | """ |
||
1373 | Allocate status quo pv rooftop generators to buildings. |
||
1374 | Parameters |
||
1375 | ----------- |
||
1376 | mastr_gdf : geopandas.GeoDataFrame |
||
1377 | GeoDataFrame containing MaStR data with geocoded locations. |
||
1378 | buildings_gdf : geopandas.GeoDataFrame |
||
1379 | GeoDataFrame containing OSM buildings data with buildings without an |
||
1380 | AGS ID dropped. |
||
1381 | Returns |
||
1382 | ------- |
||
1383 | tuple with two geopandas.GeoDataFrame s |
||
1384 | GeoDataFrame containing MaStR data allocated to building IDs. |
||
1385 | GeoDataFrame containing building data allocated to MaStR IDs. |
||
1386 | """ |
||
1387 | logger.debug("Starting allocation of status quo.") |
||
1388 | |||
1389 | q_mastr_gdf = sort_and_qcut_df(mastr_gdf, col="capacity", q=Q) |
||
1390 | q_buildings_gdf = sort_and_qcut_df(buildings_gdf, col="building_area", q=Q) |
||
1391 | |||
1392 | desagg_mastr_gdf, desagg_buildings_gdf = allocate_pv( |
||
1393 | q_mastr_gdf, q_buildings_gdf, SEED |
||
1394 | ) |
||
1395 | |||
1396 | validate_output(desagg_mastr_gdf, desagg_buildings_gdf) |
||
1397 | |||
1398 | return drop_unallocated_gens(desagg_mastr_gdf), desagg_buildings_gdf |
||
1399 | |||
1400 | |||
1401 | @timer_func |
||
1402 | def grid_districts( |
||
1403 | epsg: int, |
||
1404 | ) -> gpd.GeoDataFrame: |
||
1405 | """ |
||
1406 | Load mv grid district geo data from eGo^n Database as |
||
1407 | geopandas.GeoDataFrame. |
||
1408 | Parameters |
||
1409 | ----------- |
||
1410 | epsg : int |
||
1411 | EPSG ID to use as CRS. |
||
1412 | Returns |
||
1413 | ------- |
||
1414 | geopandas.GeoDataFrame |
||
1415 | GeoDataFrame containing mv grid district ID and geo shapes data. |
||
1416 | """ |
||
1417 | gdf = db.select_geodataframe( |
||
1418 | """ |
||
1419 | SELECT bus_id, geom |
||
1420 | FROM grid.egon_mv_grid_district |
||
1421 | ORDER BY bus_id |
||
1422 | """, |
||
1423 | index_col="bus_id", |
||
1424 | geom_col="geom", |
||
1425 | epsg=epsg, |
||
1426 | ) |
||
1427 | |||
1428 | gdf.index = gdf.index.astype(int) |
||
1429 | |||
1430 | logger.debug("Grid districts loaded.") |
||
1431 | |||
1432 | return gdf |
||
1433 | |||
1434 | |||
1435 | def scenario_data( |
||
1436 | carrier: str = "solar_rooftop", |
||
1437 | scenario: str = "eGon2035", |
||
1438 | ) -> pd.DataFrame: |
||
1439 | """ |
||
1440 | Get scenario capacity data from eGo^n Database. |
||
1441 | Parameters |
||
1442 | ----------- |
||
1443 | carrier : str |
||
1444 | Carrier type to filter table by. |
||
1445 | scenario : str |
||
1446 | Scenario to filter table by. |
||
1447 | Returns |
||
1448 | ------- |
||
1449 | geopandas.GeoDataFrame |
||
1450 | GeoDataFrame with scenario capacity data in GW. |
||
1451 | """ |
||
1452 | with db.session_scope() as session: |
||
1453 | query = session.query(EgonScenarioCapacities).filter( |
||
1454 | EgonScenarioCapacities.carrier == carrier, |
||
1455 | EgonScenarioCapacities.scenario_name == scenario, |
||
1456 | ) |
||
1457 | |||
1458 | df = pd.read_sql( |
||
1459 | query.statement, query.session.bind, index_col="index" |
||
1460 | ).sort_index() |
||
1461 | |||
1462 | logger.debug("Scenario capacity data loaded.") |
||
1463 | |||
1464 | return df |
||
1465 | |||
1466 | |||
1467 | View Code Duplication | class Vg250Lan(Base): |
|
|
|||
1468 | __tablename__ = "vg250_lan" |
||
1469 | __table_args__ = {"schema": "boundaries"} |
||
1470 | |||
1471 | id = Column(BigInteger, primary_key=True, index=True) |
||
1472 | ade = Column(BigInteger) |
||
1473 | gf = Column(BigInteger) |
||
1474 | bsg = Column(BigInteger) |
||
1475 | ars = Column(String) |
||
1476 | ags = Column(String) |
||
1477 | sdv_ars = Column(String) |
||
1478 | gen = Column(String) |
||
1479 | bez = Column(String) |
||
1480 | ibz = Column(BigInteger) |
||
1481 | bem = Column(String) |
||
1482 | nbd = Column(String) |
||
1483 | sn_l = Column(String) |
||
1484 | sn_r = Column(String) |
||
1485 | sn_k = Column(String) |
||
1486 | sn_v1 = Column(String) |
||
1487 | sn_v2 = Column(String) |
||
1488 | sn_g = Column(String) |
||
1489 | fk_s3 = Column(String) |
||
1490 | nuts = Column(String) |
||
1491 | ars_0 = Column(String) |
||
1492 | ags_0 = Column(String) |
||
1493 | wsk = Column(String) |
||
1494 | debkg_id = Column(String) |
||
1495 | rs = Column(String) |
||
1496 | sdv_rs = Column(String) |
||
1497 | rs_0 = Column(String) |
||
1498 | geometry = Column(Geometry(srid=EPSG), index=True) |
||
1499 | |||
1500 | |||
1501 | def federal_state_data(to_crs: CRS) -> gpd.GeoDataFrame: |
||
1502 | """ |
||
1503 | Get feder state data from eGo^n Database. |
||
1504 | Parameters |
||
1505 | ----------- |
||
1506 | to_crs : pyproj.crs.crs.CRS |
||
1507 | CRS to transform geometries to. |
||
1508 | Returns |
||
1509 | ------- |
||
1510 | geopandas.GeoDataFrame |
||
1511 | GeoDataFrame with federal state data. |
||
1512 | """ |
||
1513 | with db.session_scope() as session: |
||
1514 | query = session.query( |
||
1515 | Vg250Lan.id, Vg250Lan.nuts, Vg250Lan.geometry.label("geom") |
||
1516 | ) |
||
1517 | |||
1518 | gdf = gpd.read_postgis( |
||
1519 | query.statement, session.connection(), index_col="id" |
||
1520 | ).to_crs(to_crs) |
||
1521 | |||
1522 | logger.debug("Federal State data loaded.") |
||
1523 | |||
1524 | return gdf |
||
1525 | |||
1526 | |||
1527 | @timer_func |
||
1528 | def overlay_grid_districts_with_counties( |
||
1529 | mv_grid_district_gdf: gpd.GeoDataFrame, |
||
1530 | federal_state_gdf: gpd.GeoDataFrame, |
||
1531 | ) -> gpd.GeoDataFrame: |
||
1532 | """ |
||
1533 | Calculate the intersections of mv grid districts and counties. |
||
1534 | Parameters |
||
1535 | ----------- |
||
1536 | mv_grid_district_gdf : gpd.GeoDataFrame |
||
1537 | GeoDataFrame containing mv grid district ID and geo shapes data. |
||
1538 | federal_state_gdf : gpd.GeoDataFrame |
||
1539 | GeoDataFrame with federal state data. |
||
1540 | Returns |
||
1541 | ------- |
||
1542 | geopandas.GeoDataFrame |
||
1543 | GeoDataFrame containing OSM buildings data. |
||
1544 | """ |
||
1545 | logger.debug( |
||
1546 | "Calculating intersection overlay between mv grid districts and " |
||
1547 | "counties. This may take a while..." |
||
1548 | ) |
||
1549 | |||
1550 | gdf = gpd.overlay( |
||
1551 | federal_state_gdf.to_crs(mv_grid_district_gdf.crs), |
||
1552 | mv_grid_district_gdf.reset_index(), |
||
1553 | how="intersection", |
||
1554 | keep_geom_type=True, |
||
1555 | ) |
||
1556 | |||
1557 | logger.debug("Done!") |
||
1558 | |||
1559 | return gdf |
||
1560 | |||
1561 | |||
1562 | @timer_func |
||
1563 | def add_overlay_id_to_buildings( |
||
1564 | buildings_gdf: gpd.GeoDataFrame, |
||
1565 | grid_federal_state_gdf: gpd.GeoDataFrame, |
||
1566 | ) -> gpd.GeoDataFrame: |
||
1567 | """ |
||
1568 | Add information about overlay ID to buildings. |
||
1569 | Parameters |
||
1570 | ----------- |
||
1571 | buildings_gdf : geopandas.GeoDataFrame |
||
1572 | GeoDataFrame containing OSM buildings data. |
||
1573 | grid_federal_state_gdf : geopandas.GeoDataFrame |
||
1574 | GeoDataFrame with intersection shapes between counties and grid |
||
1575 | districts. |
||
1576 | Returns |
||
1577 | ------- |
||
1578 | geopandas.GeoDataFrame |
||
1579 | GeoDataFrame containing OSM buildings data with overlay ID added. |
||
1580 | """ |
||
1581 | gdf = ( |
||
1582 | buildings_gdf.to_crs(grid_federal_state_gdf.crs) |
||
1583 | .sjoin( |
||
1584 | grid_federal_state_gdf, |
||
1585 | how="left", |
||
1586 | predicate="intersects", |
||
1587 | ) |
||
1588 | .rename(columns={"index_right": "overlay_id"}) |
||
1589 | ) |
||
1590 | |||
1591 | logger.debug("Added overlay ID to OSM buildings.") |
||
1592 | |||
1593 | return gdf |
||
1594 | |||
1595 | |||
1596 | def drop_buildings_outside_grids( |
||
1597 | buildings_gdf: gpd.GeoDataFrame, |
||
1598 | ) -> gpd.GeoDataFrame: |
||
1599 | """ |
||
1600 | Drop all buildings outside of grid areas. |
||
1601 | Parameters |
||
1602 | ----------- |
||
1603 | buildings_gdf : geopandas.GeoDataFrame |
||
1604 | GeoDataFrame containing OSM buildings data. |
||
1605 | Returns |
||
1606 | ------- |
||
1607 | gepandas.GeoDataFrame |
||
1608 | GeoDataFrame containing OSM buildings data |
||
1609 | with buildings without an bus ID dropped. |
||
1610 | """ |
||
1611 | gdf = buildings_gdf.loc[~buildings_gdf.bus_id.isna()] |
||
1612 | |||
1613 | logger.debug( |
||
1614 | f"{len(buildings_gdf) - len(gdf)} " |
||
1615 | f"({(len(buildings_gdf) - len(gdf)) / len(buildings_gdf) * 100:g}%) " |
||
1616 | f"of {len(buildings_gdf)} values are outside of the grid areas " |
||
1617 | "and are therefore dropped." |
||
1618 | ) |
||
1619 | |||
1620 | return gdf |
||
1621 | |||
1622 | |||
1623 | def cap_per_bus_id( |
||
1624 | scenario: str, |
||
1625 | ) -> pd.DataFrame: |
||
1626 | """ |
||
1627 | Get table with total pv rooftop capacity per grid district. |
||
1628 | |||
1629 | Parameters |
||
1630 | ----------- |
||
1631 | scenario : str |
||
1632 | Scenario name. |
||
1633 | Returns |
||
1634 | ------- |
||
1635 | pandas.DataFrame |
||
1636 | DataFrame with total rooftop capacity per mv grid. |
||
1637 | """ |
||
1638 | targets = config.datasets()["solar_rooftop"]["targets"] |
||
1639 | |||
1640 | sql = f""" |
||
1641 | SELECT bus as bus_id, control, p_nom as capacity |
||
1642 | FROM {targets['generators']['schema']}.{targets['generators']['table']} |
||
1643 | WHERE carrier = 'solar_rooftop' |
||
1644 | AND scn_name = '{scenario}' |
||
1645 | """ |
||
1646 | # TODO: woher kommen die Slack rows??? |
||
1647 | |||
1648 | df = db.select_dataframe(sql, index_col="bus_id") |
||
1649 | |||
1650 | return df.loc[df.control != "Slack"] |
||
1651 | |||
1652 | # overlay_gdf = overlay_gdf.assign(capacity=np.nan) |
||
1653 | # |
||
1654 | # for cap, nuts in scenario_df[["capacity", "nuts"]].itertuples(index=False): |
||
1655 | # nuts_gdf = overlay_gdf.loc[overlay_gdf.nuts == nuts] |
||
1656 | # |
||
1657 | # capacity = nuts_gdf.building_area.multiply( |
||
1658 | # cap / nuts_gdf.building_area.sum() |
||
1659 | # ) |
||
1660 | # |
||
1661 | # overlay_gdf.loc[nuts_gdf.index] = overlay_gdf.loc[ |
||
1662 | # nuts_gdf.index |
||
1663 | # ].assign(capacity=capacity.multiply(conversion).to_numpy()) |
||
1664 | # |
||
1665 | # return overlay_gdf[["bus_id", "capacity"]].groupby("bus_id").sum() |
||
1666 | |||
1667 | |||
1668 | def determine_end_of_life_gens( |
||
1669 | mastr_gdf: gpd.GeoDataFrame, |
||
1670 | scenario_timestamp: pd.Timestamp, |
||
1671 | pv_rooftop_lifetime: pd.Timedelta, |
||
1672 | ) -> gpd.GeoDataFrame: |
||
1673 | """ |
||
1674 | Determine if an old PV system has reached its end of life. |
||
1675 | Parameters |
||
1676 | ----------- |
||
1677 | mastr_gdf : geopandas.GeoDataFrame |
||
1678 | GeoDataFrame containing geocoded MaStR data. |
||
1679 | scenario_timestamp : pandas.Timestamp |
||
1680 | Timestamp at which the scenario takes place. |
||
1681 | pv_rooftop_lifetime : pandas.Timedelta |
||
1682 | Average expected lifetime of PV rooftop systems. |
||
1683 | Returns |
||
1684 | ------- |
||
1685 | geopandas.GeoDataFrame |
||
1686 | GeoDataFrame containing geocoded MaStR data and info if the system |
||
1687 | has reached its end of life. |
||
1688 | """ |
||
1689 | before = mastr_gdf.capacity.sum() |
||
1690 | |||
1691 | mastr_gdf = mastr_gdf.assign( |
||
1692 | age=scenario_timestamp - mastr_gdf.start_up_date |
||
1693 | ) |
||
1694 | |||
1695 | mastr_gdf = mastr_gdf.assign( |
||
1696 | end_of_life=pv_rooftop_lifetime < mastr_gdf.age |
||
1697 | ) |
||
1698 | |||
1699 | after = mastr_gdf.loc[~mastr_gdf.end_of_life].capacity.sum() |
||
1700 | |||
1701 | logger.debug( |
||
1702 | f"Determined if pv rooftop systems reached their end of life.\nTotal " |
||
1703 | f"capacity: {before}\nActive capacity: {after}" |
||
1704 | ) |
||
1705 | |||
1706 | return mastr_gdf |
||
1707 | |||
1708 | |||
1709 | def calculate_max_pv_cap_per_building( |
||
1710 | buildings_gdf: gpd.GeoDataFrame, |
||
1711 | mastr_gdf: gpd.GeoDataFrame, |
||
1712 | pv_cap_per_sq_m: float | int, |
||
1713 | roof_factor: float | int, |
||
1714 | ) -> gpd.GeoDataFrame: |
||
1715 | """ |
||
1716 | Calculate the estimated maximum possible PV capacity per building. |
||
1717 | Parameters |
||
1718 | ----------- |
||
1719 | buildings_gdf : geopandas.GeoDataFrame |
||
1720 | GeoDataFrame containing OSM buildings data. |
||
1721 | mastr_gdf : geopandas.GeoDataFrame |
||
1722 | GeoDataFrame containing geocoded MaStR data. |
||
1723 | pv_cap_per_sq_m : float, int |
||
1724 | Average expected, installable PV capacity per square meter. |
||
1725 | roof_factor : float, int |
||
1726 | Average for PV usable roof area share. |
||
1727 | Returns |
||
1728 | ------- |
||
1729 | geopandas.GeoDataFrame |
||
1730 | GeoDataFrame containing OSM buildings data with estimated maximum PV |
||
1731 | capacity. |
||
1732 | """ |
||
1733 | gdf = ( |
||
1734 | buildings_gdf.reset_index() |
||
1735 | .rename(columns={"index": "id"}) |
||
1736 | .merge( |
||
1737 | mastr_gdf[ |
||
1738 | [ |
||
1739 | "capacity", |
||
1740 | "end_of_life", |
||
1741 | "building_id", |
||
1742 | "EinheitlicheAusrichtungUndNeigungswinkel", |
||
1743 | "Hauptausrichtung", |
||
1744 | "HauptausrichtungNeigungswinkel", |
||
1745 | ] |
||
1746 | ], |
||
1747 | how="left", |
||
1748 | left_on="id", |
||
1749 | right_on="building_id", |
||
1750 | ) |
||
1751 | .set_index("id") |
||
1752 | .drop(columns="building_id") |
||
1753 | ) |
||
1754 | |||
1755 | return gdf.assign( |
||
1756 | max_cap=gdf.building_area.multiply(roof_factor * pv_cap_per_sq_m), |
||
1757 | end_of_life=gdf.end_of_life.fillna(True).astype(bool), |
||
1758 | bus_id=gdf.bus_id.astype(int), |
||
1759 | ) |
||
1760 | |||
1761 | |||
1762 | def calculate_building_load_factor( |
||
1763 | mastr_gdf: gpd.GeoDataFrame, |
||
1764 | buildings_gdf: gpd.GeoDataFrame, |
||
1765 | rounding: int = 4, |
||
1766 | ) -> gpd.GeoDataFrame: |
||
1767 | """ |
||
1768 | Calculate the roof load factor from existing PV systems. |
||
1769 | Parameters |
||
1770 | ----------- |
||
1771 | mastr_gdf : geopandas.GeoDataFrame |
||
1772 | GeoDataFrame containing geocoded MaStR data. |
||
1773 | buildings_gdf : geopandas.GeoDataFrame |
||
1774 | GeoDataFrame containing OSM buildings data. |
||
1775 | rounding : int |
||
1776 | Rounding to use for load factor. |
||
1777 | Returns |
||
1778 | ------- |
||
1779 | geopandas.GeoDataFrame |
||
1780 | GeoDataFrame containing geocoded MaStR data with calculated load |
||
1781 | factor. |
||
1782 | """ |
||
1783 | gdf = mastr_gdf.merge( |
||
1784 | buildings_gdf[["max_cap", "building_area"]] |
||
1785 | .loc[~buildings_gdf["max_cap"].isna()] |
||
1786 | .reset_index(), |
||
1787 | how="left", |
||
1788 | left_on="building_id", |
||
1789 | right_on="id", |
||
1790 | ).set_index("id") |
||
1791 | |||
1792 | return gdf.assign(load_factor=(gdf.capacity / gdf.max_cap).round(rounding)) |
||
1793 | |||
1794 | |||
1795 | def get_probability_for_property( |
||
1796 | mastr_gdf: gpd.GeoDataFrame, |
||
1797 | cap_range: tuple[int | float, int | float], |
||
1798 | prop: str, |
||
1799 | ) -> tuple[np.array, np.array]: |
||
1800 | """ |
||
1801 | Calculate the probability of the different options of a property of the |
||
1802 | existing PV plants. |
||
1803 | Parameters |
||
1804 | ----------- |
||
1805 | mastr_gdf : geopandas.GeoDataFrame |
||
1806 | GeoDataFrame containing geocoded MaStR data. |
||
1807 | cap_range : tuple(int, int) |
||
1808 | Capacity range of PV plants to look at. |
||
1809 | prop : str |
||
1810 | Property to calculate probabilities for. String needs to be in columns |
||
1811 | of mastr_gdf. |
||
1812 | Returns |
||
1813 | ------- |
||
1814 | tuple |
||
1815 | numpy.array |
||
1816 | Unique values of property. |
||
1817 | numpy.array |
||
1818 | Probabilties per unique value. |
||
1819 | """ |
||
1820 | cap_range_gdf = mastr_gdf.loc[ |
||
1821 | (mastr_gdf.capacity > cap_range[0]) |
||
1822 | & (mastr_gdf.capacity <= cap_range[1]) |
||
1823 | ] |
||
1824 | |||
1825 | if prop == "load_factor": |
||
1826 | cap_range_gdf = cap_range_gdf.loc[cap_range_gdf[prop] <= 1] |
||
1827 | |||
1828 | count = Counter( |
||
1829 | cap_range_gdf[prop].loc[ |
||
1830 | ~cap_range_gdf[prop].isna() |
||
1831 | & ~cap_range_gdf[prop].isnull() |
||
1832 | & ~(cap_range_gdf[prop] == "None") |
||
1833 | ] |
||
1834 | ) |
||
1835 | |||
1836 | values = np.array(list(count.keys())) |
||
1837 | probabilities = np.fromiter(count.values(), dtype=float) |
||
1838 | probabilities = probabilities / np.sum(probabilities) |
||
1839 | |||
1840 | return values, probabilities |
||
1841 | |||
1842 | |||
1843 | @timer_func |
||
1844 | def probabilities( |
||
1845 | mastr_gdf: gpd.GeoDataFrame, |
||
1846 | cap_ranges: list[tuple[int | float, int | float]] | None = None, |
||
1847 | properties: list[str] | None = None, |
||
1848 | ) -> dict: |
||
1849 | """ |
||
1850 | Calculate the probability of the different options of properties of the |
||
1851 | existing PV plants. |
||
1852 | Parameters |
||
1853 | ----------- |
||
1854 | mastr_gdf : geopandas.GeoDataFrame |
||
1855 | GeoDataFrame containing geocoded MaStR data. |
||
1856 | cap_ranges : list(tuple(int, int)) |
||
1857 | List of capacity ranges to distinguish between. The first tuple should |
||
1858 | start with a zero and the last one should end with infinite. |
||
1859 | properties : list(str) |
||
1860 | List of properties to calculate probabilities for. Strings need to be |
||
1861 | in columns of mastr_gdf. |
||
1862 | Returns |
||
1863 | ------- |
||
1864 | dict |
||
1865 | Dictionary with values and probabilities per capacity range. |
||
1866 | """ |
||
1867 | if cap_ranges is None: |
||
1868 | cap_ranges = [ |
||
1869 | (0, 30), |
||
1870 | (30, 100), |
||
1871 | (100, float("inf")), |
||
1872 | ] |
||
1873 | if properties is None: |
||
1874 | properties = [ |
||
1875 | "EinheitlicheAusrichtungUndNeigungswinkel", |
||
1876 | "Hauptausrichtung", |
||
1877 | "HauptausrichtungNeigungswinkel", |
||
1878 | "load_factor", |
||
1879 | ] |
||
1880 | |||
1881 | prob_dict = {} |
||
1882 | |||
1883 | for cap_range in cap_ranges: |
||
1884 | prob_dict[cap_range] = { |
||
1885 | "values": {}, |
||
1886 | "probabilities": {}, |
||
1887 | } |
||
1888 | |||
1889 | for prop in properties: |
||
1890 | v, p = get_probability_for_property( |
||
1891 | mastr_gdf, |
||
1892 | cap_range, |
||
1893 | prop, |
||
1894 | ) |
||
1895 | |||
1896 | prob_dict[cap_range]["values"][prop] = v |
||
1897 | prob_dict[cap_range]["probabilities"][prop] = p |
||
1898 | |||
1899 | return prob_dict |
||
1900 | |||
1901 | |||
1902 | def cap_share_per_cap_range( |
||
1903 | mastr_gdf: gpd.GeoDataFrame, |
||
1904 | cap_ranges: list[tuple[int | float, int | float]] | None = None, |
||
1905 | ) -> dict[tuple[int | float, int | float], float]: |
||
1906 | """ |
||
1907 | Calculate the share of PV capacity from the total PV capacity within |
||
1908 | capacity ranges. |
||
1909 | Parameters |
||
1910 | ----------- |
||
1911 | mastr_gdf : geopandas.GeoDataFrame |
||
1912 | GeoDataFrame containing geocoded MaStR data. |
||
1913 | cap_ranges : list(tuple(int, int)) |
||
1914 | List of capacity ranges to distinguish between. The first tuple should |
||
1915 | start with a zero and the last one should end with infinite. |
||
1916 | Returns |
||
1917 | ------- |
||
1918 | dict |
||
1919 | Dictionary with share of PV capacity from the total PV capacity within |
||
1920 | capacity ranges. |
||
1921 | """ |
||
1922 | if cap_ranges is None: |
||
1923 | cap_ranges = [ |
||
1924 | (0, 30), |
||
1925 | (30, 100), |
||
1926 | (100, float("inf")), |
||
1927 | ] |
||
1928 | |||
1929 | cap_share_dict = {} |
||
1930 | |||
1931 | total_cap = mastr_gdf.capacity.sum() |
||
1932 | |||
1933 | for cap_range in cap_ranges: |
||
1934 | cap_share = ( |
||
1935 | mastr_gdf.loc[ |
||
1936 | (mastr_gdf.capacity > cap_range[0]) |
||
1937 | & (mastr_gdf.capacity <= cap_range[1]) |
||
1938 | ].capacity.sum() |
||
1939 | / total_cap |
||
1940 | ) |
||
1941 | |||
1942 | cap_share_dict[cap_range] = cap_share |
||
1943 | |||
1944 | return cap_share_dict |
||
1945 | |||
1946 | |||
1947 | def mean_load_factor_per_cap_range( |
||
1948 | mastr_gdf: gpd.GeoDataFrame, |
||
1949 | cap_ranges: list[tuple[int | float, int | float]] | None = None, |
||
1950 | ) -> dict[tuple[int | float, int | float], float]: |
||
1951 | """ |
||
1952 | Calculate the mean roof load factor per capacity range from existing PV |
||
1953 | plants. |
||
1954 | Parameters |
||
1955 | ----------- |
||
1956 | mastr_gdf : geopandas.GeoDataFrame |
||
1957 | GeoDataFrame containing geocoded MaStR data. |
||
1958 | cap_ranges : list(tuple(int, int)) |
||
1959 | List of capacity ranges to distinguish between. The first tuple should |
||
1960 | start with a zero and the last one should end with infinite. |
||
1961 | Returns |
||
1962 | ------- |
||
1963 | dict |
||
1964 | Dictionary with mean roof load factor per capacity range. |
||
1965 | """ |
||
1966 | if cap_ranges is None: |
||
1967 | cap_ranges = [ |
||
1968 | (0, 30), |
||
1969 | (30, 100), |
||
1970 | (100, float("inf")), |
||
1971 | ] |
||
1972 | |||
1973 | load_factor_dict = {} |
||
1974 | |||
1975 | for cap_range in cap_ranges: |
||
1976 | load_factor = mastr_gdf.loc[ |
||
1977 | (mastr_gdf.load_factor <= 1) |
||
1978 | & (mastr_gdf.capacity > cap_range[0]) |
||
1979 | & (mastr_gdf.capacity <= cap_range[1]) |
||
1980 | ].load_factor.mean() |
||
1981 | |||
1982 | load_factor_dict[cap_range] = load_factor |
||
1983 | |||
1984 | return load_factor_dict |
||
1985 | |||
1986 | |||
1987 | def building_area_range_per_cap_range( |
||
1988 | mastr_gdf: gpd.GeoDataFrame, |
||
1989 | cap_ranges: list[tuple[int | float, int | float]] | None = None, |
||
1990 | min_building_size: int | float = 10.0, |
||
1991 | upper_quantile: float = 0.95, |
||
1992 | lower_quantile: float = 0.05, |
||
1993 | ) -> dict[tuple[int | float, int | float], tuple[int | float, int | float]]: |
||
1994 | """ |
||
1995 | Estimate normal building area range per capacity range. |
||
1996 | Calculate the mean roof load factor per capacity range from existing PV |
||
1997 | plants. |
||
1998 | Parameters |
||
1999 | ----------- |
||
2000 | mastr_gdf : geopandas.GeoDataFrame |
||
2001 | GeoDataFrame containing geocoded MaStR data. |
||
2002 | cap_ranges : list(tuple(int, int)) |
||
2003 | List of capacity ranges to distinguish between. The first tuple should |
||
2004 | start with a zero and the last one should end with infinite. |
||
2005 | min_building_size : int, float |
||
2006 | Minimal building size to consider for PV plants. |
||
2007 | upper_quantile : float |
||
2008 | Upper quantile to estimate maximum building size per capacity range. |
||
2009 | lower_quantile : float |
||
2010 | Lower quantile to estimate minimum building size per capacity range. |
||
2011 | Returns |
||
2012 | ------- |
||
2013 | dict |
||
2014 | Dictionary with estimated normal building area range per capacity |
||
2015 | range. |
||
2016 | """ |
||
2017 | if cap_ranges is None: |
||
2018 | cap_ranges = [ |
||
2019 | (0, 30), |
||
2020 | (30, 100), |
||
2021 | (100, float("inf")), |
||
2022 | ] |
||
2023 | |||
2024 | building_area_range_dict = {} |
||
2025 | |||
2026 | n_ranges = len(cap_ranges) |
||
2027 | |||
2028 | for count, cap_range in enumerate(cap_ranges): |
||
2029 | cap_range_gdf = mastr_gdf.loc[ |
||
2030 | (mastr_gdf.capacity > cap_range[0]) |
||
2031 | & (mastr_gdf.capacity <= cap_range[1]) |
||
2032 | ] |
||
2033 | |||
2034 | if count == 0: |
||
2035 | building_area_range_dict[cap_range] = ( |
||
2036 | min_building_size, |
||
2037 | cap_range_gdf.building_area.quantile(upper_quantile), |
||
2038 | ) |
||
2039 | elif count == n_ranges - 1: |
||
2040 | building_area_range_dict[cap_range] = ( |
||
2041 | cap_range_gdf.building_area.quantile(lower_quantile), |
||
2042 | float("inf"), |
||
2043 | ) |
||
2044 | else: |
||
2045 | building_area_range_dict[cap_range] = ( |
||
2046 | cap_range_gdf.building_area.quantile(lower_quantile), |
||
2047 | cap_range_gdf.building_area.quantile(upper_quantile), |
||
2048 | ) |
||
2049 | |||
2050 | values = list(building_area_range_dict.values()) |
||
2051 | |||
2052 | building_area_range_normed_dict = {} |
||
2053 | |||
2054 | for count, (cap_range, (min_area, max_area)) in enumerate( |
||
2055 | building_area_range_dict.items() |
||
2056 | ): |
||
2057 | if count == 0: |
||
2058 | building_area_range_normed_dict[cap_range] = ( |
||
2059 | min_area, |
||
2060 | np.mean((values[count + 1][0], max_area)), |
||
2061 | ) |
||
2062 | elif count == n_ranges - 1: |
||
2063 | building_area_range_normed_dict[cap_range] = ( |
||
2064 | np.mean((values[count - 1][1], min_area)), |
||
2065 | max_area, |
||
2066 | ) |
||
2067 | else: |
||
2068 | building_area_range_normed_dict[cap_range] = ( |
||
2069 | np.mean((values[count - 1][1], min_area)), |
||
2070 | np.mean((values[count + 1][0], max_area)), |
||
2071 | ) |
||
2072 | |||
2073 | return building_area_range_normed_dict |
||
2074 | |||
2075 | |||
2076 | @timer_func |
||
2077 | def desaggregate_pv_in_mv_grid( |
||
2078 | buildings_gdf: gpd.GeoDataFrame, |
||
2079 | pv_cap: float | int, |
||
2080 | **kwargs, |
||
2081 | ) -> gpd.GeoDataFrame: |
||
2082 | """ |
||
2083 | Desaggregate PV capacity on buildings within a given grid district. |
||
2084 | Parameters |
||
2085 | ----------- |
||
2086 | buildings_gdf : geopandas.GeoDataFrame |
||
2087 | GeoDataFrame containing buildings within the grid district. |
||
2088 | pv_cap : float, int |
||
2089 | PV capacity to desaggregate. |
||
2090 | Other Parameters |
||
2091 | ----------- |
||
2092 | prob_dict : dict |
||
2093 | Dictionary with values and probabilities per capacity range. |
||
2094 | cap_share_dict : dict |
||
2095 | Dictionary with share of PV capacity from the total PV capacity within |
||
2096 | capacity ranges. |
||
2097 | building_area_range_dict : dict |
||
2098 | Dictionary with estimated normal building area range per capacity |
||
2099 | range. |
||
2100 | load_factor_dict : dict |
||
2101 | Dictionary with mean roof load factor per capacity range. |
||
2102 | seed : int |
||
2103 | Seed to use for random operations with NumPy and pandas. |
||
2104 | pv_cap_per_sq_m : float, int |
||
2105 | Average expected, installable PV capacity per square meter. |
||
2106 | Returns |
||
2107 | ------- |
||
2108 | geopandas.GeoDataFrame |
||
2109 | GeoDataFrame containing OSM building data with desaggregated PV |
||
2110 | plants. |
||
2111 | """ |
||
2112 | bus_id = int(buildings_gdf.bus_id.iat[0]) |
||
2113 | |||
2114 | rng = default_rng(seed=kwargs["seed"]) |
||
2115 | random_state = RandomState(seed=kwargs["seed"]) |
||
2116 | |||
2117 | results_df = pd.DataFrame(columns=buildings_gdf.columns) |
||
2118 | |||
2119 | for cap_range, share in kwargs["cap_share_dict"].items(): |
||
2120 | pv_cap_range = pv_cap * share |
||
2121 | |||
2122 | b_area_min, b_area_max = kwargs["building_area_range_dict"][cap_range] |
||
2123 | |||
2124 | cap_range_buildings_gdf = buildings_gdf.loc[ |
||
2125 | ~buildings_gdf.index.isin(results_df.index) |
||
2126 | & (buildings_gdf.building_area > b_area_min) |
||
2127 | & (buildings_gdf.building_area <= b_area_max) |
||
2128 | ] |
||
2129 | |||
2130 | mean_load_factor = kwargs["load_factor_dict"][cap_range] |
||
2131 | cap_range_buildings_gdf = cap_range_buildings_gdf.assign( |
||
2132 | mean_cap=cap_range_buildings_gdf.max_cap * mean_load_factor, |
||
2133 | load_factor=np.nan, |
||
2134 | capacity=np.nan, |
||
2135 | ) |
||
2136 | |||
2137 | total_mean_cap = cap_range_buildings_gdf.mean_cap.sum() |
||
2138 | |||
2139 | if total_mean_cap == 0: |
||
2140 | logger.warning( |
||
2141 | f"There are no matching roof for capacity range {cap_range} " |
||
2142 | f"kW in grid {bus_id}. Using all buildings as fallback." |
||
2143 | ) |
||
2144 | |||
2145 | cap_range_buildings_gdf = buildings_gdf.loc[ |
||
2146 | ~buildings_gdf.index.isin(results_df.index) |
||
2147 | ] |
||
2148 | |||
2149 | if len(cap_range_buildings_gdf) == 0: |
||
2150 | logger.warning( |
||
2151 | "There are no roofes available for capacity range " |
||
2152 | f"{cap_range} kW in grid {bus_id}. Allowing dual use." |
||
2153 | ) |
||
2154 | cap_range_buildings_gdf = buildings_gdf.copy() |
||
2155 | |||
2156 | cap_range_buildings_gdf = cap_range_buildings_gdf.assign( |
||
2157 | mean_cap=cap_range_buildings_gdf.max_cap * mean_load_factor, |
||
2158 | load_factor=np.nan, |
||
2159 | capacity=np.nan, |
||
2160 | ) |
||
2161 | |||
2162 | total_mean_cap = cap_range_buildings_gdf.mean_cap.sum() |
||
2163 | |||
2164 | elif total_mean_cap < pv_cap_range: |
||
2165 | logger.warning( |
||
2166 | f"Average roof utilization of the roof area in grid {bus_id} " |
||
2167 | f"and capacity range {cap_range} kW is not sufficient. The " |
||
2168 | "roof utilization will be above average." |
||
2169 | ) |
||
2170 | |||
2171 | frac = max( |
||
2172 | pv_cap_range / total_mean_cap, |
||
2173 | 1 / len(cap_range_buildings_gdf), |
||
2174 | ) |
||
2175 | |||
2176 | samples_gdf = cap_range_buildings_gdf.sample( |
||
2177 | frac=min(1, frac), |
||
2178 | random_state=random_state, |
||
2179 | ) |
||
2180 | |||
2181 | cap_range_dict = kwargs["prob_dict"][cap_range] |
||
2182 | |||
2183 | values_dict = cap_range_dict["values"] |
||
2184 | p_dict = cap_range_dict["probabilities"] |
||
2185 | |||
2186 | load_factors = rng.choice( |
||
2187 | a=values_dict["load_factor"], |
||
2188 | size=len(samples_gdf), |
||
2189 | p=p_dict["load_factor"], |
||
2190 | ) |
||
2191 | |||
2192 | samples_gdf = samples_gdf.assign( |
||
2193 | load_factor=load_factors, |
||
2194 | capacity=( |
||
2195 | samples_gdf.building_area |
||
2196 | * load_factors |
||
2197 | * kwargs["pv_cap_per_sq_m"] |
||
2198 | ).clip(lower=0.4), |
||
2199 | ) |
||
2200 | |||
2201 | missing_factor = pv_cap_range / samples_gdf.capacity.sum() |
||
2202 | |||
2203 | samples_gdf = samples_gdf.assign( |
||
2204 | capacity=(samples_gdf.capacity * missing_factor), |
||
2205 | load_factor=(samples_gdf.load_factor * missing_factor), |
||
2206 | ) |
||
2207 | |||
2208 | assert np.isclose( |
||
2209 | samples_gdf.capacity.sum(), |
||
2210 | pv_cap_range, |
||
2211 | rtol=1e-03, |
||
2212 | ), f"{samples_gdf.capacity.sum()} != {pv_cap_range}" |
||
2213 | |||
2214 | results_df = pd.concat( |
||
2215 | [ |
||
2216 | results_df, |
||
2217 | samples_gdf, |
||
2218 | ], |
||
2219 | ) |
||
2220 | |||
2221 | total_missing_factor = pv_cap / results_df.capacity.sum() |
||
2222 | |||
2223 | results_df = results_df.assign( |
||
2224 | capacity=(results_df.capacity * total_missing_factor), |
||
2225 | ) |
||
2226 | |||
2227 | assert np.isclose( |
||
2228 | results_df.capacity.sum(), |
||
2229 | pv_cap, |
||
2230 | rtol=1e-03, |
||
2231 | ), f"{results_df.capacity.sum()} != {pv_cap}" |
||
2232 | |||
2233 | return gpd.GeoDataFrame( |
||
2234 | results_df, |
||
2235 | crs=samples_gdf.crs, |
||
2236 | geometry="geom", |
||
2237 | ) |
||
2238 | |||
2239 | |||
2240 | @timer_func |
||
2241 | def desaggregate_pv( |
||
2242 | buildings_gdf: gpd.GeoDataFrame, |
||
2243 | cap_df: pd.DataFrame, |
||
2244 | **kwargs, |
||
2245 | ) -> gpd.GeoDataFrame: |
||
2246 | """ |
||
2247 | Desaggregate PV capacity on buildings within a given grid district. |
||
2248 | Parameters |
||
2249 | ----------- |
||
2250 | buildings_gdf : geopandas.GeoDataFrame |
||
2251 | GeoDataFrame containing OSM buildings data. |
||
2252 | cap_df : pandas.DataFrame |
||
2253 | DataFrame with total rooftop capacity per mv grid. |
||
2254 | Other Parameters |
||
2255 | ----------- |
||
2256 | prob_dict : dict |
||
2257 | Dictionary with values and probabilities per capacity range. |
||
2258 | cap_share_dict : dict |
||
2259 | Dictionary with share of PV capacity from the total PV capacity within |
||
2260 | capacity ranges. |
||
2261 | building_area_range_dict : dict |
||
2262 | Dictionary with estimated normal building area range per capacity |
||
2263 | range. |
||
2264 | load_factor_dict : dict |
||
2265 | Dictionary with mean roof load factor per capacity range. |
||
2266 | seed : int |
||
2267 | Seed to use for random operations with NumPy and pandas. |
||
2268 | pv_cap_per_sq_m : float, int |
||
2269 | Average expected, installable PV capacity per square meter. |
||
2270 | Returns |
||
2271 | ------- |
||
2272 | geopandas.GeoDataFrame |
||
2273 | GeoDataFrame containing OSM building data with desaggregated PV |
||
2274 | plants. |
||
2275 | """ |
||
2276 | allocated_buildings_gdf = buildings_gdf.loc[~buildings_gdf.end_of_life] |
||
2277 | |||
2278 | building_bus_ids = set(buildings_gdf.bus_id) |
||
2279 | cap_bus_ids = set(cap_df.index) |
||
2280 | |||
2281 | logger.debug( |
||
2282 | f"Bus IDs from buildings: {len(building_bus_ids)}\nBus IDs from " |
||
2283 | f"capacity: {len(cap_bus_ids)}" |
||
2284 | ) |
||
2285 | |||
2286 | if len(building_bus_ids) > len(cap_bus_ids): |
||
2287 | missing = building_bus_ids - cap_bus_ids |
||
2288 | else: |
||
2289 | missing = cap_bus_ids - building_bus_ids |
||
2290 | |||
2291 | logger.debug(str(missing)) |
||
2292 | |||
2293 | bus_ids = np.intersect1d(list(building_bus_ids), list(cap_bus_ids)) |
||
2294 | |||
2295 | # assert set(buildings_gdf.bus_id.unique()) == set(cap_df.index) |
||
2296 | |||
2297 | for bus_id in bus_ids: |
||
2298 | buildings_grid_gdf = buildings_gdf.loc[buildings_gdf.bus_id == bus_id] |
||
2299 | |||
2300 | pv_installed_gdf = buildings_grid_gdf.loc[ |
||
2301 | ~buildings_grid_gdf.end_of_life |
||
2302 | ] |
||
2303 | |||
2304 | pv_installed = pv_installed_gdf.capacity.sum() |
||
2305 | |||
2306 | pot_buildings_gdf = buildings_grid_gdf.drop( |
||
2307 | index=pv_installed_gdf.index |
||
2308 | ) |
||
2309 | |||
2310 | if len(pot_buildings_gdf) == 0: |
||
2311 | logger.error( |
||
2312 | f"In grid {bus_id} there are no potential buildings to " |
||
2313 | f"allocate PV capacity to. The grid is skipped. This message " |
||
2314 | f"should only appear doing test runs with few buildings." |
||
2315 | ) |
||
2316 | |||
2317 | continue |
||
2318 | |||
2319 | pv_target = cap_df.at[bus_id, "capacity"] * 1000 |
||
2320 | |||
2321 | logger.debug(f"pv_target: {pv_target}") |
||
2322 | |||
2323 | pv_missing = pv_target - pv_installed |
||
2324 | |||
2325 | if pv_missing <= 0: |
||
2326 | logger.warning( |
||
2327 | f"In grid {bus_id} there is more PV installed " |
||
2328 | f"({pv_installed: g} kW) in status Quo than allocated within " |
||
2329 | f"the scenario ({pv_target: g} kW). " |
||
2330 | f"No new generators are added." |
||
2331 | ) |
||
2332 | |||
2333 | continue |
||
2334 | |||
2335 | if pot_buildings_gdf.max_cap.sum() < pv_missing: |
||
2336 | logger.error( |
||
2337 | f"In grid {bus_id} there is less PV potential (" |
||
2338 | f"{pot_buildings_gdf.max_cap.sum():g} kW) than allocated PV " |
||
2339 | f"capacity ({pv_missing:g} kW). The average roof utilization " |
||
2340 | f"will be very high." |
||
2341 | ) |
||
2342 | |||
2343 | gdf = desaggregate_pv_in_mv_grid( |
||
2344 | buildings_gdf=pot_buildings_gdf, |
||
2345 | pv_cap=pv_missing, |
||
2346 | **kwargs, |
||
2347 | ) |
||
2348 | |||
2349 | logger.debug(f"New cap in grid {bus_id}: {gdf.capacity.sum()}") |
||
2350 | logger.debug(f"Installed cap in grid {bus_id}: {pv_installed}") |
||
2351 | logger.debug( |
||
2352 | f"Total cap in grid {bus_id}: {gdf.capacity.sum() + pv_installed}" |
||
2353 | ) |
||
2354 | |||
2355 | if not np.isclose( |
||
2356 | gdf.capacity.sum() + pv_installed, pv_target, rtol=1e-3 |
||
2357 | ): |
||
2358 | logger.warning( |
||
2359 | f"The desired capacity and actual capacity in grid {bus_id} " |
||
2360 | f"differ.\n" |
||
2361 | f"Desired cap: {pv_target}\nActual cap: " |
||
2362 | f"{gdf.capacity.sum() + pv_installed}" |
||
2363 | ) |
||
2364 | |||
2365 | pre_cap = allocated_buildings_gdf.capacity.sum() |
||
2366 | new_cap = gdf.capacity.sum() |
||
2367 | |||
2368 | allocated_buildings_gdf = pd.concat( |
||
2369 | [ |
||
2370 | allocated_buildings_gdf, |
||
2371 | gdf, |
||
2372 | ] |
||
2373 | ) |
||
2374 | |||
2375 | total_cap = allocated_buildings_gdf.capacity.sum() |
||
2376 | |||
2377 | assert np.isclose(pre_cap + new_cap, total_cap) |
||
2378 | |||
2379 | logger.debug("Desaggregated scenario.") |
||
2380 | logger.debug(f"Scenario capacity: {cap_df.capacity.sum(): g}") |
||
2381 | logger.debug( |
||
2382 | f"Generator capacity: " |
||
2383 | f"{allocated_buildings_gdf.capacity.sum() / 1000: g}" |
||
2384 | ) |
||
2385 | |||
2386 | return gpd.GeoDataFrame( |
||
2387 | allocated_buildings_gdf, |
||
2388 | crs=gdf.crs, |
||
2389 | geometry="geom", |
||
2390 | ) |
||
2391 | |||
2392 | |||
2393 | @timer_func |
||
2394 | def add_buildings_meta_data( |
||
2395 | buildings_gdf: gpd.GeoDataFrame, |
||
2396 | prob_dict: dict, |
||
2397 | seed: int, |
||
2398 | ) -> gpd.GeoDataFrame: |
||
2399 | """ |
||
2400 | Randomly add additional metadata to desaggregated PV plants. |
||
2401 | Parameters |
||
2402 | ----------- |
||
2403 | buildings_gdf : geopandas.GeoDataFrame |
||
2404 | GeoDataFrame containing OSM buildings data with desaggregated PV |
||
2405 | plants. |
||
2406 | prob_dict : dict |
||
2407 | Dictionary with values and probabilities per capacity range. |
||
2408 | seed : int |
||
2409 | Seed to use for random operations with NumPy and pandas. |
||
2410 | Returns |
||
2411 | ------- |
||
2412 | geopandas.GeoDataFrame |
||
2413 | GeoDataFrame containing OSM building data with desaggregated PV |
||
2414 | plants. |
||
2415 | """ |
||
2416 | rng = default_rng(seed=seed) |
||
2417 | buildings_gdf = buildings_gdf.reset_index().rename( |
||
2418 | columns={ |
||
2419 | "index": "building_id", |
||
2420 | } |
||
2421 | ) |
||
2422 | |||
2423 | for (min_cap, max_cap), cap_range_prob_dict in prob_dict.items(): |
||
2424 | cap_range_gdf = buildings_gdf.loc[ |
||
2425 | (buildings_gdf.capacity >= min_cap) |
||
2426 | & (buildings_gdf.capacity < max_cap) |
||
2427 | ] |
||
2428 | |||
2429 | for key, values in cap_range_prob_dict["values"].items(): |
||
2430 | if key == "load_factor": |
||
2431 | continue |
||
2432 | |||
2433 | gdf = cap_range_gdf.loc[ |
||
2434 | cap_range_gdf[key].isna() |
||
2435 | | cap_range_gdf[key].isnull() |
||
2436 | | (cap_range_gdf[key] == "None") |
||
2437 | ] |
||
2438 | |||
2439 | key_vals = rng.choice( |
||
2440 | a=values, |
||
2441 | size=len(gdf), |
||
2442 | p=cap_range_prob_dict["probabilities"][key], |
||
2443 | ) |
||
2444 | |||
2445 | buildings_gdf.loc[gdf.index, key] = key_vals |
||
2446 | |||
2447 | return buildings_gdf |
||
2448 | |||
2449 | |||
2450 | def add_voltage_level( |
||
2451 | buildings_gdf: gpd.GeoDataFrame, |
||
2452 | ) -> gpd.GeoDataFrame: |
||
2453 | """ |
||
2454 | Get voltage level data from mastr table and assign to units. Infer missing |
||
2455 | values derived from generator capacity to the power plants. |
||
2456 | |||
2457 | Parameters |
||
2458 | ----------- |
||
2459 | buildings_gdf : geopandas.GeoDataFrame |
||
2460 | GeoDataFrame containing OSM buildings data with desaggregated PV |
||
2461 | plants. |
||
2462 | Returns |
||
2463 | ------- |
||
2464 | geopandas.GeoDataFrame |
||
2465 | GeoDataFrame containing OSM building data with voltage level per |
||
2466 | generator. |
||
2467 | """ |
||
2468 | |||
2469 | View Code Duplication | def voltage_levels(p: float) -> int: |
|
2470 | if p <= 100: |
||
2471 | return 7 |
||
2472 | elif p <= 200: |
||
2473 | return 6 |
||
2474 | elif p <= 5500: |
||
2475 | return 5 |
||
2476 | elif p <= 20000: |
||
2477 | return 4 |
||
2478 | elif p <= 120000: |
||
2479 | return 3 |
||
2480 | return 1 |
||
2481 | |||
2482 | # Join mastr table |
||
2483 | with db.session_scope() as session: |
||
2484 | query = session.query( |
||
2485 | EgonPowerPlantsPv.gens_id, |
||
2486 | EgonPowerPlantsPv.voltage_level, |
||
2487 | ) |
||
2488 | voltage_levels_df = pd.read_sql( |
||
2489 | query.statement, query.session.bind, index_col=None |
||
2490 | ) |
||
2491 | buildings_gdf = buildings_gdf.merge( |
||
2492 | voltage_levels_df, |
||
2493 | left_on="gens_id", |
||
2494 | right_on="gens_id", |
||
2495 | how="left", |
||
2496 | ) |
||
2497 | |||
2498 | # Infer missing values |
||
2499 | mask = buildings_gdf.voltage_level.isna() |
||
2500 | buildings_gdf.loc[mask, "voltage_level"] = buildings_gdf.loc[ |
||
2501 | mask |
||
2502 | ].capacity.apply(voltage_levels) |
||
2503 | |||
2504 | return buildings_gdf |
||
2505 | |||
2506 | |||
2507 | def add_start_up_date( |
||
2508 | buildings_gdf: gpd.GeoDataFrame, |
||
2509 | start: pd.Timestamp, |
||
2510 | end: pd.Timestamp, |
||
2511 | seed: int, |
||
2512 | ): |
||
2513 | """ |
||
2514 | Randomly and linear add start-up date to new pv generators. |
||
2515 | Parameters |
||
2516 | ---------- |
||
2517 | buildings_gdf : geopandas.GeoDataFrame |
||
2518 | GeoDataFrame containing OSM buildings data with desaggregated PV |
||
2519 | plants. |
||
2520 | start : pandas.Timestamp |
||
2521 | Minimum Timestamp to use. |
||
2522 | end : pandas.Timestamp |
||
2523 | Maximum Timestamp to use. |
||
2524 | seed : int |
||
2525 | Seed to use for random operations with NumPy and pandas. |
||
2526 | Returns |
||
2527 | ------- |
||
2528 | geopandas.GeoDataFrame |
||
2529 | GeoDataFrame containing OSM buildings data with start-up date added. |
||
2530 | """ |
||
2531 | rng = default_rng(seed=seed) |
||
2532 | |||
2533 | date_range = pd.date_range(start=start, end=end, freq="1D") |
||
2534 | |||
2535 | return buildings_gdf.assign( |
||
2536 | start_up_date=rng.choice(date_range, size=len(buildings_gdf)) |
||
2537 | ) |
||
2538 | |||
2539 | |||
2540 | @timer_func |
||
2541 | def allocate_scenarios( |
||
2542 | mastr_gdf: gpd.GeoDataFrame, |
||
2543 | valid_buildings_gdf: gpd.GeoDataFrame, |
||
2544 | last_scenario_gdf: gpd.GeoDataFrame, |
||
2545 | scenario: str, |
||
2546 | ): |
||
2547 | """ |
||
2548 | Desaggregate and allocate scenario pv rooftop ramp-ups onto buildings. |
||
2549 | Parameters |
||
2550 | ---------- |
||
2551 | mastr_gdf : geopandas.GeoDataFrame |
||
2552 | GeoDataFrame containing geocoded MaStR data. |
||
2553 | valid_buildings_gdf : geopandas.GeoDataFrame |
||
2554 | GeoDataFrame containing OSM buildings data. |
||
2555 | last_scenario_gdf : geopandas.GeoDataFrame |
||
2556 | GeoDataFrame containing OSM buildings matched with pv generators from |
||
2557 | temporally preceding scenario. |
||
2558 | scenario : str |
||
2559 | Scenario to desaggrgate and allocate. |
||
2560 | Returns |
||
2561 | ------- |
||
2562 | tuple |
||
2563 | geopandas.GeoDataFrame |
||
2564 | GeoDataFrame containing OSM buildings matched with pv generators. |
||
2565 | pandas.DataFrame |
||
2566 | DataFrame containing pv rooftop capacity per grid id. |
||
2567 | """ |
||
2568 | cap_per_bus_id_df = cap_per_bus_id(scenario) |
||
2569 | |||
2570 | logger.debug( |
||
2571 | f"cap_per_bus_id_df total capacity: {cap_per_bus_id_df.capacity.sum()}" |
||
2572 | ) |
||
2573 | |||
2574 | last_scenario_gdf = determine_end_of_life_gens( |
||
2575 | last_scenario_gdf, |
||
2576 | SCENARIO_TIMESTAMP[scenario], |
||
2577 | PV_ROOFTOP_LIFETIME, |
||
2578 | ) |
||
2579 | |||
2580 | buildings_gdf = calculate_max_pv_cap_per_building( |
||
2581 | valid_buildings_gdf, |
||
2582 | last_scenario_gdf, |
||
2583 | PV_CAP_PER_SQ_M, |
||
2584 | ROOF_FACTOR, |
||
2585 | ) |
||
2586 | |||
2587 | mastr_gdf = calculate_building_load_factor( |
||
2588 | mastr_gdf, |
||
2589 | buildings_gdf, |
||
2590 | ) |
||
2591 | |||
2592 | probabilities_dict = probabilities( |
||
2593 | mastr_gdf, |
||
2594 | cap_ranges=CAP_RANGES, |
||
2595 | ) |
||
2596 | |||
2597 | cap_share_dict = cap_share_per_cap_range( |
||
2598 | mastr_gdf, |
||
2599 | cap_ranges=CAP_RANGES, |
||
2600 | ) |
||
2601 | |||
2602 | load_factor_dict = mean_load_factor_per_cap_range( |
||
2603 | mastr_gdf, |
||
2604 | cap_ranges=CAP_RANGES, |
||
2605 | ) |
||
2606 | |||
2607 | building_area_range_dict = building_area_range_per_cap_range( |
||
2608 | mastr_gdf, |
||
2609 | cap_ranges=CAP_RANGES, |
||
2610 | min_building_size=MIN_BUILDING_SIZE, |
||
2611 | upper_quantile=UPPER_QUNATILE, |
||
2612 | lower_quantile=LOWER_QUANTILE, |
||
2613 | ) |
||
2614 | |||
2615 | allocated_buildings_gdf = desaggregate_pv( |
||
2616 | buildings_gdf=buildings_gdf, |
||
2617 | cap_df=cap_per_bus_id_df, |
||
2618 | prob_dict=probabilities_dict, |
||
2619 | cap_share_dict=cap_share_dict, |
||
2620 | building_area_range_dict=building_area_range_dict, |
||
2621 | load_factor_dict=load_factor_dict, |
||
2622 | seed=SEED, |
||
2623 | pv_cap_per_sq_m=PV_CAP_PER_SQ_M, |
||
2624 | ) |
||
2625 | |||
2626 | allocated_buildings_gdf = allocated_buildings_gdf.assign(scenario=scenario) |
||
2627 | |||
2628 | meta_buildings_gdf = frame_to_numeric( |
||
2629 | add_buildings_meta_data( |
||
2630 | allocated_buildings_gdf, |
||
2631 | probabilities_dict, |
||
2632 | SEED, |
||
2633 | ) |
||
2634 | ) |
||
2635 | |||
2636 | return ( |
||
2637 | add_start_up_date( |
||
2638 | meta_buildings_gdf, |
||
2639 | start=last_scenario_gdf.start_up_date.max(), |
||
2640 | end=SCENARIO_TIMESTAMP[scenario], |
||
2641 | seed=SEED, |
||
2642 | ), |
||
2643 | cap_per_bus_id_df, |
||
2644 | ) |
||
2645 | |||
2646 | |||
2647 | class EgonPowerPlantPvRoofBuildingScenario(Base): |
||
2648 | __tablename__ = "egon_power_plants_pv_roof_building" |
||
2649 | __table_args__ = {"schema": "supply"} |
||
2650 | |||
2651 | index = Column(Integer, primary_key=True, index=True) |
||
2652 | scenario = Column(String) |
||
2653 | bus_id = Column(Integer, nullable=True) |
||
2654 | building_id = Column(Integer) |
||
2655 | gens_id = Column(String, nullable=True) |
||
2656 | capacity = Column(Float) |
||
2657 | einheitliche_ausrichtung_und_neigungswinkel = Column(Float) |
||
2658 | hauptausrichtung = Column(String) |
||
2659 | hauptausrichtung_neigungswinkel = Column(String) |
||
2660 | voltage_level = Column(Integer) |
||
2661 | weather_cell_id = Column(Integer) |
||
2662 | |||
2663 | |||
2664 | def create_scenario_table(buildings_gdf): |
||
2665 | """Create mapping table pv_unit <-> building for scenario""" |
||
2666 | EgonPowerPlantPvRoofBuildingScenario.__table__.drop( |
||
2667 | bind=engine, checkfirst=True |
||
2668 | ) |
||
2669 | EgonPowerPlantPvRoofBuildingScenario.__table__.create( |
||
2670 | bind=engine, checkfirst=True |
||
2671 | ) |
||
2672 | |||
2673 | buildings_gdf.rename(columns=COLS_TO_RENAME).assign( |
||
2674 | capacity=buildings_gdf.capacity.div(10**3) # kW -> MW |
||
2675 | )[COLS_TO_EXPORT].reset_index().to_sql( |
||
2676 | name=EgonPowerPlantPvRoofBuildingScenario.__table__.name, |
||
2677 | schema=EgonPowerPlantPvRoofBuildingScenario.__table__.schema, |
||
2678 | con=db.engine(), |
||
2679 | if_exists="append", |
||
2680 | index=False, |
||
2681 | ) |
||
2682 | |||
2683 | |||
2684 | def geocode_mastr_data(): |
||
2685 | """ |
||
2686 | Read PV rooftop data from MaStR CSV |
||
2687 | TODO: the source will be replaced as soon as the MaStR data is available |
||
2688 | in DB. |
||
2689 | """ |
||
2690 | mastr_df = mastr_data( |
||
2691 | MASTR_INDEX_COL, |
||
2692 | MASTR_RELEVANT_COLS, |
||
2693 | MASTR_DTYPES, |
||
2694 | MASTR_PARSE_DATES, |
||
2695 | ) |
||
2696 | |||
2697 | clean_mastr_df = clean_mastr_data( |
||
2698 | mastr_df, |
||
2699 | max_realistic_pv_cap=MAX_REALISTIC_PV_CAP, |
||
2700 | min_realistic_pv_cap=MIN_REALISTIC_PV_CAP, |
||
2701 | seed=SEED, |
||
2702 | rounding=ROUNDING, |
||
2703 | ) |
||
2704 | |||
2705 | geocoding_df = geocoding_data(clean_mastr_df) |
||
2706 | |||
2707 | ratelimiter = geocoder(USER_AGENT, MIN_DELAY_SECONDS) |
||
2708 | |||
2709 | geocode_gdf = geocode_data(geocoding_df, ratelimiter, EPSG) |
||
2710 | |||
2711 | create_geocoded_table(geocode_gdf) |
||
2712 | |||
2713 | |||
2714 | def add_weather_cell_id(buildings_gdf: gpd.GeoDataFrame) -> gpd.GeoDataFrame: |
||
2715 | sql = """ |
||
2716 | SELECT building_id, zensus_population_id |
||
2717 | FROM boundaries.egon_map_zensus_mvgd_buildings |
||
2718 | """ |
||
2719 | |||
2720 | buildings_gdf = buildings_gdf.merge( |
||
2721 | right=db.select_dataframe(sql).drop_duplicates(subset="building_id"), |
||
2722 | how="left", |
||
2723 | on="building_id", |
||
2724 | ) |
||
2725 | |||
2726 | sql = """ |
||
2727 | SELECT zensus_population_id, w_id as weather_cell_id |
||
2728 | FROM boundaries.egon_map_zensus_weather_cell |
||
2729 | """ |
||
2730 | |||
2731 | buildings_gdf = buildings_gdf.merge( |
||
2732 | right=db.select_dataframe(sql).drop_duplicates( |
||
2733 | subset="zensus_population_id" |
||
2734 | ), |
||
2735 | how="left", |
||
2736 | on="zensus_population_id", |
||
2737 | ) |
||
2738 | |||
2739 | if buildings_gdf.weather_cell_id.isna().any(): |
||
2740 | missing = buildings_gdf.loc[ |
||
2741 | buildings_gdf.weather_cell_id.isna() |
||
2742 | ].building_id.tolist() |
||
2743 | |||
2744 | raise ValueError( |
||
2745 | f"Following buildings don't have a weather cell id: {missing}" |
||
2746 | ) |
||
2747 | |||
2748 | return buildings_gdf |
||
2749 | |||
2750 | |||
2751 | def add_bus_ids_sq(buildings_gdf: gpd.GeoDataFrame,) -> gpd.GeoDataFrame: |
||
2752 | """Add bus ids for status_quo units |
||
2753 | |||
2754 | Parameters |
||
2755 | ----------- |
||
2756 | buildings_gdf : geopandas.GeoDataFrame |
||
2757 | GeoDataFrame containing OSM buildings data with desaggregated PV |
||
2758 | plants. |
||
2759 | Returns |
||
2760 | ------- |
||
2761 | geopandas.GeoDataFrame |
||
2762 | GeoDataFrame containing OSM building data with bus_id per |
||
2763 | generator. |
||
2764 | """ |
||
2765 | grid_districts_gdf = grid_districts(EPSG) |
||
2766 | |||
2767 | mask = buildings_gdf.scenario == "status_quo" |
||
2768 | buildings_gdf.loc[mask, "bus_id"] = ( |
||
2769 | buildings_gdf.loc[mask] |
||
2770 | .sjoin(grid_districts_gdf, how="left") |
||
2771 | .index_right |
||
2772 | ) |
||
2773 | |||
2774 | return buildings_gdf |
||
2775 | |||
2776 | |||
2777 | def pv_rooftop_to_buildings(): |
||
2778 | """Main script, executed as task""" |
||
2779 | |||
2780 | mastr_gdf = load_mastr_data() |
||
2781 | |||
2782 | buildings_gdf = load_building_data() |
||
2783 | |||
2784 | desagg_mastr_gdf, desagg_buildings_gdf = allocate_to_buildings( |
||
2785 | mastr_gdf, buildings_gdf |
||
2786 | ) |
||
2787 | |||
2788 | all_buildings_gdf = ( |
||
2789 | desagg_mastr_gdf.assign(scenario="status_quo") |
||
2790 | .reset_index() |
||
2791 | .rename(columns={"geometry": "geom", "EinheitMastrNummer": "gens_id"}) |
||
2792 | ) |
||
2793 | |||
2794 | scenario_buildings_gdf = all_buildings_gdf.copy() |
||
2795 | |||
2796 | cap_per_bus_id_df = pd.DataFrame() |
||
2797 | |||
2798 | for scenario in SCENARIOS: |
||
2799 | logger.debug(f"Desaggregating scenario {scenario}.") |
||
2800 | ( |
||
2801 | scenario_buildings_gdf, |
||
2802 | cap_per_bus_id_scenario_df, |
||
2803 | ) = allocate_scenarios( # noqa: F841 |
||
2804 | desagg_mastr_gdf, |
||
2805 | desagg_buildings_gdf, |
||
2806 | scenario_buildings_gdf, |
||
2807 | scenario, |
||
2808 | ) |
||
2809 | |||
2810 | all_buildings_gdf = gpd.GeoDataFrame( |
||
2811 | pd.concat( |
||
2812 | [all_buildings_gdf, scenario_buildings_gdf], ignore_index=True |
||
2813 | ), |
||
2814 | crs=scenario_buildings_gdf.crs, |
||
2815 | geometry="geom", |
||
2816 | ) |
||
2817 | |||
2818 | cap_per_bus_id_df = pd.concat( |
||
2819 | [cap_per_bus_id_df, cap_per_bus_id_scenario_df] |
||
2820 | ) |
||
2821 | |||
2822 | # add weather cell |
||
2823 | all_buildings_gdf = add_weather_cell_id(all_buildings_gdf) |
||
2824 | |||
2825 | # add bus IDs for status quo scenario |
||
2826 | all_buildings_gdf = add_bus_ids_sq(all_buildings_gdf) |
||
2827 | |||
2828 | # export scenario |
||
2829 | create_scenario_table(add_voltage_level(all_buildings_gdf)) |
||
2830 |