Source code for message_ix_models.model.transport.operator

""":mod:`genno` operators for MESSAGEix-Transport."""

import logging
import re
from functools import partial, reduce
from itertools import product
from operator import gt, le, lt
from typing import (
    TYPE_CHECKING,
    Any,
    Dict,
    Hashable,
    List,
    Mapping,
    Optional,
    Sequence,
    Set,
    Tuple,
    cast,
)

import genno
import numpy as np
import pandas as pd
import xarray as xr
from genno import Computer, KeySeq, Operator, quote
from genno.operator import apply_units, rename_dims
from genno.testing import assert_qty_allclose, assert_units
from sdmx.model.v21 import Code

from message_ix_models import ScenarioInfo
from message_ix_models.model.structure import get_codelist, get_codes
from message_ix_models.project.navigate import T35_POLICY
from message_ix_models.report.operator import compound_growth
from message_ix_models.report.util import as_quantity
from message_ix_models.util import (
    MappingAdapter,
    broadcast,
    datetime_now_with_tz,
    nodes_ex_world,
    show_versions,
)

from .config import Config

if TYPE_CHECKING:
    from genno.types import AnyQuantity
    from message_ix import Scenario
    from xarray.core.types import Dims

    import message_ix_models.model.transport.factor
    from message_ix_models import Context

log = logging.getLogger(__name__)

__all__ = [
    "base_model_data_header",
    "base_shares",
    "broadcast_advance",
    "broadcast_y_yv_ya",
    "cost",
    "distance_ldv",
    "distance_nonldv",
    "dummy_prices",
    "extend_y",
    "factor_fv",
    "factor_input",
    "factor_pdt",
    "groups_iea_eweb",
    "iea_eei_fv",
    "indexers_n_cd",
    "indexers_usage",
    "input_commodity_level",
    "logit",
    "max",
    "min",
    "merge_data",
    "nodes_ex_world",  # Re-export from message_ix_models.util TODO do this upstream
    "nodes_world_agg",
    "price_units",
    "quantity_from_config",
    "relabel2",
    "share_weight",
    "smooth",
    "transport_check",
    "transport_data",
    "votm",
]


[docs]def base_model_data_header(scenario: "Scenario", *, name: str) -> Dict[str, str]: """Return a header comment for writing out base model data.""" versions = "\n\n".join(show_versions().split("\n\n")[:2]) return dict( header_comment=f"""`{name}` parameter data for MESSAGEix-GLOBIOM. Generated: {datetime_now_with_tz().isoformat()} from: ixmp://{scenario.platform.name}/{scenario.url} using: {versions} """ )
[docs]def base_shares( base: "AnyQuantity", nodes: List[str], techs: List[str], y: List[int] ) -> "AnyQuantity": """Return base mode shares. The mode shares are read from a file at :file:`data/transport/{regions}/mode-shares/{name}.csv`, where `name` is from the configuration key ``mode-share:``, and `region` uses :func:`.path_fallback`. Labels on the t (technology) dimension must match the ``demand modes:`` from the configuration. If the data lack the n (node, spatial) and/or y (time) dimensions, they are broadcast over these. """ from genno.operator import aggregate, sum from numpy import allclose # Check: ensure values sum to 1 tmp = sum(base, dimensions=["t"]) check = allclose(tmp.to_series().values, 1.0) if not check: log.warning( "Sum across modes ≠ 1.0; will rescale:\n" + (tmp[tmp != 1.0].to_string()) ) result = base / tmp else: result = base assert allclose(sum(result, dimensions=["t"]).to_series().values, 1.0) # Aggregate extra modes that do not appear in the data extra_modes = set(result.coords["t"].data) - set(techs) if extra_modes == {"OTHER ROAD"}: # Add "OTHER ROAD" to "LDV" groups = {t: [t] for t in techs} groups["LDV"].append("OTHER ROAD") result = aggregate(result, groups=dict(t=groups), keep=False) elif len(extra_modes): raise NotImplementedError(f"Extra mode(s) t={extra_modes}") missing = cast(Set[Hashable], set("nty")) - set(result.dims) if len(missing): log.info(f"Broadcast base mode shares with dims {base.dims} over {missing}") coords = [("n", nodes), ("t", techs), ("y", y)] result = base * genno.Quantity(xr.DataArray(1.0, coords=coords), units="") return result
[docs]def broadcast_advance(data: "AnyQuantity", y0: int, config: dict) -> "AnyQuantity": """Broadcast ADVANCE `data` from native `n` coords to :py:`config["regions"]`.""" from genno.operator import sum assert "R12" == config["regions"], "ADVANCE data mapping only for R12 regions" # Create a quantity for broadcasting df = pd.DataFrame( [ ["ASIA", 0.1, "R12_RCPA"], ["ASIA", 0.5 - 0.1, "R12_PAS"], ["CHN", 1.0, "R12_CHN"], ["EU", 0.1, "R12_EEU"], ["EU", 0.9, "R12_WEU"], ["IND", 1.0, "R12_SAS"], ["LAM", 1.0, "R12_LAM"], ["MAF", 0.5, "R12_AFR"], ["MAF", 0.5, "R12_MEA"], ["OECD90", 0.08, "R12_PAO"], ["REF", 1.0, "R12_FSU"], ["USA", 1.0, "R12_NAM"], ], columns=["n", "value", "n_new"], ) bcast = genno.Quantity(df.set_index(["n", "n_new"])["value"]) check = data.sel(n="World", y=y0, drop=True) # - Multiply by `bcast`, adding a new dimension "n_new". # - Sum on "n" and drop that dimension. # - Rename "n_new" to "n". result = ( (data.sel(y=y0, drop=True) * bcast) .pipe(sum, dimensions=["n"]) .pipe(rename_dims, {"n_new": "n"}) ) # Ensure the total across the new `n` coords still matches the world total assert_qty_allclose(check, sum(result, dimensions=["n"]), rtol=5e-2) return result
[docs]def broadcast_y_yv_ya(y: List[int], y_model: List[int]) -> "AnyQuantity": """Return a quantity for broadcasting y to (yv, ya). This is distinct from :attr:`.ScenarioInfo.ya_ya`, because it omits all :math:`y^V < y_0`. """ dims = ["y", "yv", "ya"] series = ( pd.DataFrame(product(y, y_model), columns=dims[1:]) .query("ya >= yv") .assign(value=1.0, y=lambda df: df["yv"]) .set_index(dims)["value"] ) return genno.Quantity(series)
[docs]def cost( price: "AnyQuantity", gdp_ppp_cap: "AnyQuantity", whours: "AnyQuantity", speeds: "AnyQuantity", votm: "AnyQuantity", y: List[int], ) -> "AnyQuantity": """Calculate cost of transport [money / distance]. Calculated from two components: 1. The inherent price of the mode. 2. Value of time, in turn from: 1. a value of time multiplier (`votm`), 2. the wage rate per hour (`gdp_ppp_cap` / `whours`), and 3. the travel time per unit distance (1 / `speeds`). """ from genno.operator import add # NB for some reason, the 'y' dimension of result becomes `float`, rather than # `int`, in this step # FIXME genno does not handle units here correctly using "+" instead of add() return add(price, (gdp_ppp_cap * votm) / (speeds * whours)).sel(y=y)
[docs]def distance_ldv(config: dict) -> "AnyQuantity": """Return annual driving distance per LDV. - Regions other than R11_NAM have M/F values in same proportion to their A value as in NAM """ # Load from config.yaml result = as_quantity(config["transport"].ldv_activity) * as_quantity( config["transport"].factor["activity"]["ldv"] ) result.name = "ldv distance" return result
#: Mapping from technology names appearing in the IEA EEI data to those in #: MESSAGEix-Transport. EEI_TECH_MAP = { "Buses": "BUS", "Cars/light trucks": "LDV", "Freight trains": "freight rail", "Freight trucks": "freight truck", "Motorcycles": "2W", "Passenger trains": "RAIL", }
[docs]def distance_nonldv(context: "Context") -> "AnyQuantity": """Return annual travel distance per vehicle for non-LDV transport modes.""" # FIXME Remove this type exclusion; added only to merge #549 import message_ix_models.tools.iea.eei # type: ignore # noqa: F401 from message_ix_models.tools import exo_data log.warning( "distance_nonldv() currently returns a sum, rather than weighted average. Use" "with caution." ) c = Computer() source_kw = dict(measure="Vehicle use", aggregate=True) keys = exo_data.prepare_computer(context, c, "IEA EEI", source_kw) ks = KeySeq(keys[0]) c.add(ks[0], "select", ks.base, indexers={"SECTOR": "transport"}, drop=True) c.add(ks[1], "rename", ks[0], quote({"Mode/vehicle type": "t"})) # Replace IEA EEI technology codes with MESSAGEix-Transport ones c.add(ks[2], "relabel", ks[1], labels=dict(t=EEI_TECH_MAP)) # Ensure compatible dimensionality and convert units c.add(ks[3], "convert_units", ks[2], units="Mm / vehicle / year") c.add(ks[4], "rename_dims", ks[3], quote({"n": "nl"})) # Execute the calculation result = c.get(ks[4]) # Select the latest year. # TODO check whether coverage varies by year; if so, then fill-forward or # extrapolate y_m1 = result.coords["y"].data[-1] log.info(f"Return data for y={y_m1}") return result.sel(y=y_m1, drop=True)
[docs]def dummy_prices(gdp: "AnyQuantity") -> "AnyQuantity": # Commodity prices: all equal to 0.1 # Same coords/shape as `gdp`, but with c="transport" coords = [(dim, item.data) for dim, item in gdp.coords.items()] coords.append(("c", ["transport"])) shape = list(len(c[1]) for c in coords) return genno.Quantity( xr.DataArray(np.full(shape, 0.1), coords=coords), units="USD / km" )
[docs]def extend_y(qty: "AnyQuantity", y: List[int]) -> "AnyQuantity": """Extend `qty` along the "y" dimension to cover `y`.""" y_ = set(y) # Subset of `y` appearing in `qty` y_qty = sorted(set(qty.to_series().reset_index()["y"].unique()) & y_) # Subset of `target_years` to fill forward from the last period in `qty` y_to_fill = sorted(filter(partial(lt, y_qty[-1]), y_)) log.info(f"{qty.name}: extend from {y_qty[-1]}{y_to_fill}") # Map existing labels to themselves, and missing labels to the last existing one y_map = [(y, y) for y in y_qty] + [(y_qty[-1], y) for y in y_to_fill] # - Forward-fill *within* `qty` existing values. # - Use message_ix_models MappingAdapter to do the rest. return MappingAdapter({"y": y_map})(qty.ffill("y")) # type: ignore [attr-defined]
[docs]def factor_fv(n: List[str], y: List[int], config: dict) -> "AnyQuantity": """Scaling factor for freight activity. If :attr:`.Config.project` is :data:`ScenarioFlags.ACT`, the value declines from 1.0 at the first `y` to 0.865 (reduction of 13.5%) at y=2050, then constant thereafter. Otherwise, the value is 1.0 for every (`n`, `y`). """ # Empty data frame df = pd.DataFrame(columns=["value"], index=pd.Index(y, name="y")) # Default value df.iloc[0, :] = 1.0 # NAVIGATE T3.5 "act" demand-side scenario if T35_POLICY.ACT & config["transport"].project["navigate"]: years = list(filter(lambda y: y <= 2050, y)) df.loc[years, "value"] = np.interp(years, [y[0], 2050], [1.0, 0.865]) # - Fill all values forward from the latest. # - Convert to long format. # - Broadcast over all nodes `n`. # - Set dimensions as index. return genno.Quantity( df.infer_objects() .ffill() .reset_index() .assign(n=None) .pipe(broadcast, n=n) .set_index(["n", "y"])["value"], units="", )
[docs]def factor_input( y: List[int], t: List[Code], t_agg: Dict, config: dict ) -> "AnyQuantity": """Scaling factor for ``input`` (energy intensity of activity). If :attr:`.Config.project` is :data:`ScenarioFlags.TEC`, the value declines from 1.0 at the first `y` to 0.865 (reduction of 13.5%) at y=2050, then constant thereafter. Otherwise, the value is 1.0 for every (`t`, `y`). The return value includes ``y`` from 2010 onwards. """ def _not_disutility(tech): return tech.eval_annotation("is-disutility") is None techs = list(filter(_not_disutility, t)) # Empty data frame df = pd.DataFrame( columns=pd.Index(map(str, techs), name="t"), index=pd.Index(filter(partial(le, 2010), y), name="y"), ) # Default value df.iloc[0, :] = 1.0 # NAVIGATE T3.5 "tec" demand-side scenario if T35_POLICY.TEC & config["transport"].project["navigate"]: years = list(filter(partial(gt, 2050), df.index)) # Prepare a dictionary mapping technologies to their respective EI improvement # rates t_groups = t_agg["t"] value = {} for group, v in { "2W": 1.5, "BUS": 1.5, "LDV": 1.5, "freight truck": 2.0, "AIR": 1.3, }.items(): value.update({t: 1 - (v / 100.0) for t in t_groups[group]}) # Apply the rates, or 1.0 if none set for a particular technology for t_ in map(str, techs): df.loc[years, t_] = value.get(t_, 1.0) qty = genno.Quantity( df.infer_objects().fillna(1.0).reset_index().set_index("y").stack() ) return compound_growth(qty, "y")
[docs]def factor_pdt(n: List[str], y: List[int], t: List[str], config: dict) -> "AnyQuantity": """Scaling factor for passenger activity. When :attr:`.Config.scenarios` includes :attr:`ScenarioFlags.ACT` (i.e. NAVIGATE Task 3.5, demand-side scenario "act"), the value of 0.8 is specified for LDV, 2050, and all regions. This function implements this as a linear decrease between the first model period (currently 2020) and that point. Otherwise, the value is 1.0 for every (`n`, `t`, `y`). """ # Empty data frame df = pd.DataFrame(columns=t, index=pd.Index(y, name="y")) # Set 1.0 (no scaling) for first period df.iloc[0, :] = 1.0 # Handle particular scenarios if T35_POLICY.ACT & config["transport"].project["navigate"]: # NAVIGATE T3.5 "act" demand-side scenario years = list(filter(lambda y: y <= 2050, y)) df.loc[years, "LDV"] = np.interp(years, [y[0], 2050], [1.0, 0.8]) # - Fill all values forward from the latest. # - Convert to long format. # - Broadcast over all nodes `n`. # - Set dimensions as index. return genno.Quantity( df.infer_objects() .ffill() .reset_index() .melt(id_vars="y", var_name="t") .assign(n=None) .pipe(broadcast, n=n) .set_index(["n", "y", "t"])["value"], units="", )
[docs]def factor_ssp( config: dict, nodes: List[str], years: List[int], *others: List, info: "message_ix_models.model.transport.factor.Factor", extra_dims: Optional[Sequence[str]] = None, ) -> "AnyQuantity": """Return a scaling factor for an SSP realization.""" kw = dict(n=nodes, y=years, scenario=config["transport"].ssp) for dim, labels in zip(extra_dims or (), others): kw[dim] = labels return info.quantify(**kw)
Groups = Dict[str, Dict[str, List[str]]]
[docs]def groups_iea_eweb(technologies: List[Code]) -> Tuple[Groups, Groups, Dict]: """Structure for calibration to IEA Extended World Energy Balances (EWEB). Returns 3 sets of groups: 1. Groups for aggregating the EWEB data. In particular: - Labels for IEA ``product`` are aggregated to labels for MESSAGEix-Transport ``commodity``. - Labels for IEA ``flow`` are selected 1:1 *and* aggregated to a flow named "transport". 2. Groups for aggregating MESSAGEix-Transport data. In particular: - Labels for MESSAGEix-Transport transport modes (|t| dimension) are aggregated to labels for IEA ``flow``. 3. Indexers for *dis* aggregating computed scaling factors; that is, reversing (2). """ g0: Groups = dict(flow={}, product={}) g1: Groups = dict(t={}) g2: Dict = dict(t=[], t_new=[]) # Add groups from base model commodity code list: # - IEA product list → MESSAGE commodity (e.g. "lightoil") # - IEA flow list → MESSAGE technology group (e.g. "transport") for c in get_codelist("commodity"): if products := c.eval_annotation(id="iea-eweb-product"): g0["product"][c.id] = products if flows := c.eval_annotation(id="iea-eweb-flow"): g0["flow"][c.id] = flows # Add groups from MESSAGEix-Transport technology code list for t in technologies: if flows := t.eval_annotation(id="iea-eweb-flow"): target = flows[0] if len(flows) == 1 else t.id g0["flow"][target] = flows # Append the mode name, for instance "AIR" g1["t"].setdefault(target, []).append(t.id) # # Append the name of individual technologies for this mode # g1["t"][target].extend(map(lambda c: c.id, t.child)) g2["t"].append(target) g2["t_new"].append(t.id) g2["t"] = xr.DataArray(g2.pop("t"), coords=[("t_new", g2.pop("t_new"))]) return g0, g1, g2
[docs]def input_commodity_level(t: List[Code], default_level=None) -> "AnyQuantity": """Return a Quantity for broadcasting dimension (t) to (c, l) for ``input``. .. todo:: This essentially replaces :func:`.transport.util.input_commodity_level`, and is much faster. Replace usage of the other function with this one, then remove the other. """ c_info = get_codes("commodity") # Map each `tech` to a `commodity` and `level` data = [] for tech in t: # Retrieve the "input" annotation for this technology input_ = tech.eval_annotation("input") # Retrieve the code for this commodity try: # Commodity ID commodity = input_["commodity"] c_code = c_info[c_info.index(commodity)] except (KeyError, ValueError, TypeError): # TypeError: input_ is None # KeyError: "commodity" not in the annotation # ValueError: `commodity` not in c_info continue # Level, in order of precedence: # 1. Technology-specific input level from `t_code`. # 2. Default level for the commodity from `c_code`. # 3. `default_level` argument to this function. level = input_.get("level") or c_code.eval_annotation("level") or default_level data.append((tech.id, commodity, level)) idx = pd.MultiIndex.from_frame(pd.DataFrame(data, columns=["t", "c", "l"])) s = pd.Series(1.0, index=idx) return genno.Quantity(s)
[docs]def logit( x: "AnyQuantity", k: "AnyQuantity", lamda: "AnyQuantity", y: List[int], dim: str ) -> "AnyQuantity": r"""Compute probabilities for a logit random utility model. The choice probabilities have the form: .. math:: Pr(i) = \frac{k_j x_j ^{\lambda_j}} {\sum_{\forall i \in D} k_i x_i ^{\lambda_i}} \forall j \in D …where :math:`D` is the dimension named by the `dim` argument. All other dimensions are broadcast automatically. """ # Systematic utility u = (k * x**lamda).sel(y=y) # commented: for debugging # u.to_csv("u.csv") # Logit probability return u / u.sum(dim)
[docs]def max( qty: "AnyQuantity", dim: "Dims" = None, *, skipna: Optional[bool] = None, keep_attrs: Optional[bool] = None, **kwargs: Any, ) -> "AnyQuantity": """Like :meth:`xarray.DataArray.max`.""" assert skipna is keep_attrs is None and 0 == len(kwargs), NotImplementedError # FIXME This is AttrSeries only return qty.groupby(level=dim).max() # type: ignore
[docs]def min( qty: "AnyQuantity", dim: "Dims" = None, *, skipna: Optional[bool] = None, keep_attrs: Optional[bool] = None, **kwargs: Any, ) -> "AnyQuantity": """Like :meth:`xarray.DataArray.min`.""" assert skipna is keep_attrs is None and 0 == len(kwargs), NotImplementedError # FIXME This is AttrSeries only return qty.groupby(level=dim).min() # type: ignore
[docs]def merge_data( *others: Mapping[Hashable, pd.DataFrame], ) -> Dict[Hashable, pd.DataFrame]: """Slightly modified from message_ix_models.util. .. todo: move upstream or merge functionality with :func:`message_ix_models.util.merge_data`. """ keys: Set[Hashable] = reduce(lambda x, y: x | y.keys(), others, set()) return {k: pd.concat([o.get(k, None) for o in others]) for k in keys}
[docs]def iea_eei_fv(name: str, config: Dict) -> "AnyQuantity": """Returns base-year demand for freight from IEA EEI, with dimensions n-c-y.""" from message_ix_models.tools.iea import eei result = eei.as_quantity(name, config["regions"]) # type: ignore [attr-defined] ym1 = result.coords["y"].data[-1] log.info(f"Use y={ym1} data for base-year freight transport activity") assert set("nyt") == set(result.dims) return result.sel(y=ym1, t="Total freight transport", drop=True)
[docs]def indexers_n_cd(config: Dict) -> Dict[str, xr.DataArray]: """Indexers for selecting (`n`, `census_division`) → `n`. Based on :attr:`.Config.node_to_census_division`. """ n_cd_map = config["transport"].node_to_census_division n, cd = zip(*n_cd_map.items()) return dict( n=xr.DataArray(list(n), dims="n"), census_division=xr.DataArray(list(cd), dims="n"), )
[docs]def indexers_usage(technologies: List[Code]) -> Dict: """Indexers for replacing LDV `t` and `cg` with `t_new` for usage technologies.""" labels: Dict[str, List[str]] = dict(cg=[], t=[], t_new=[]) for t in technologies: if not t.eval_annotation("is-disutility"): continue t_base, *_, cg = t.id.split() labels["t"].append(t_base) labels["cg"].append(cg) labels["t_new"].append(t.id) return { "cg": xr.DataArray(labels["cg"], coords=[("t_new", labels["t_new"])]), "t": xr.DataArray(labels["t"], coords=[("t_new", labels["t_new"])]), }
[docs]def nodes_world_agg(config, dim: Hashable = "nl") -> Dict[Hashable, Mapping]: """Mapping to aggregate e.g. nl="World" from values for child nodes of "World". This mapping should be used with :func:`.genno.operator.aggregate`, giving the argument ``keep=False``. It includes 1:1 mapping from each region name to itself. .. todo:: move upstream, to :mod:`message_ix_models`. """ from message_ix_models.model.structure import get_codes result = {} for n in get_codes(f"node/{config['regions']}"): # "World" node should have no parent and some children. Countries (from # pycountry) that are omitted from a mapping have neither parent nor children. if len(n.child) and n.parent is None: name = str(n) # FIXME Remove. This is a hack to suit the legacy reporting, which expects # global aggregates at *_GLB rather than "World". new_name = f"{config['regions']}_GLB" log.info(f"Aggregates for {n!r} will be labelled {new_name!r}") name = new_name # Global total as aggregate of child nodes result = {name: list(map(str, n.child))} # Also add "no-op" aggregates e.g. "R12_AFR" is the sum of ["R12_AFR"] result.update({c: [c] for c in map(str, n.child)}) return {dim: result} raise RuntimeError("Failed to identify the World node")
[docs]def price_units(qty: "AnyQuantity") -> "AnyQuantity": """Forcibly adjust price units, if necessary.""" target = "USD_2010 / km" if not qty.units.is_compatible_with(target): log.warning(f"Adjust price units from {qty.units} to {target}") return apply_units(qty, target)
[docs]def quantity_from_config( config: dict, name: str, dimensionality: Optional[Dict] = None ) -> "AnyQuantity": if dimensionality: raise NotImplementedError result = getattr(config["transport"], name) if not isinstance(result, genno.Quantity): result = as_quantity(result) return result
[docs]def relabel2(qty: "AnyQuantity", new_dims: dict): """Replace dimensions with new ones using label templates. .. todo:: Choose a more descriptive name. """ from collections import defaultdict from genno.operator import select result = qty # Iterate over new dimensions for which templates are provided for new_dim, template in new_dims.items(): if new_dim in qty.dims: # pragma: no cover print(qty.coords) raise NotImplementedError( f"Replace existing dimension {new_dim} in {qty.dims}" ) # Identify 1 or more source dimension(s) in the template expression # TODO improve the regex or use another method that can handle e.g. "{func(t)}" source_dims = re.findall(r"{([^\.}]*)", template) # Resulting labels labels = defaultdict(list) # Iterate over the Cartesian product of coords on the `source_dims` for values in product(*[qty.coords[d].data for d in source_dims]): _locals = dict() # Locals for eval() for d, v in zip(source_dims, values): _locals[d] = v labels[d].append(v) # Construct the label for the `new_dim` by formatting the template string labels[new_dim].append(eval(f"f'{template}'", None, _locals)) # Convert matched lists of labels to xarray selectors selectors = { d: xr.DataArray(labels[d], coords=[(new_dim, labels[new_dim])]) for d in source_dims } result = select(result, selectors) return result
[docs]def share_weight( share: "AnyQuantity", gdp: "AnyQuantity", cost: "AnyQuantity", lamda: "AnyQuantity", t_modes: List[str], y: List[int], config: dict, ) -> "AnyQuantity": """Calculate mode share weights. - In the base year (:py:`y[0]`), the weights for each |n| are as given in `share`. - In the convergence year (:attr:`.Config.year_convergence`, via :py:`config["transport"]`), the weights are between the same-node base year mode shares and the mean of the base-year mode shares in 0 or more reference nodes given by the mapping :attr:`Config.share_weight_convergence`. - The interpolation between these points is given by the ratio :math:`k` between the same-node convergence-year GDP PPP per capita (`gdp`) and the reference node(s mean) base-year GDP PPP per capita. - If no reference nodes are given, the values converge towards equal weights for each of `t_modes`, with a fixed parameter :math:`k = 1/3`. - Values for the years between the base year and the convergence year are interpolated. Parameters ---------- gdp : GDP per capita in purchasing power parity. Returns ------- Quantity With dimensions :math:`(n, t, y)`: |n| matching `gdp_ppp_cap; :math:`t` per `t_modes`, and |y| per `y`. """ from builtins import min # Extract info from arguments cfg: Config = config["transport"] nodes = sorted(gdp.coords["n"].data) years = list(filter(lambda year: year <= cfg.year_convergence, y)) # Empty container for share weights weight = xr.DataArray(np.nan, coords=[("n", nodes), ("y", years), ("t", t_modes)]) # Selectors # A scalar induces xarray but not genno <= 1.21 to drop y0: Dict[Any, Any] = dict(y=y[0]) y0_ = dict(y=[y[0]]) # Do not drop yC: Dict[Any, Any] = dict(y=cfg.year_convergence) # Weights in y0 for all modes and nodes # NB here and below, with Python 3.9 one could do: dict(t=modes, n=nodes) | y0 idx = dict(t=t_modes, n=nodes, **y0) w0 = share.sel(idx) / (cost.sel(idx).sel(c="transport", drop=True) ** lamda) # Normalize to 1 across modes w0 = w0 / w0.sum("t") # Insert into `weight` *_, weight.loc[y0_] = xr.align(weight, xr.DataArray.from_series(w0.to_series())) # Weights at the convergence year, yC for node in nodes: # Retrieve reference nodes: a set of 0+ nodes to converge towards ref_nodes = cfg.share_weight_convergence[node] # Indexers _1 = dict(n=node, **yC) # Same node, convergence year _2 = dict(n=ref_nodes, **y0) # Reference node(s), base year if ref_nodes: # Ratio between this node's GDP in yC and the mean of the reference nodes' # GDP values in y0. Maximum 1.0. k = min( (gdp.sel(_1) / (gdp.sel(_2).sum() / float(len(ref_nodes)))).item(), 1.0 ) # As k tends to 1, converge towards the mean of the reference nodes' share # weights in y0/base shares. target = weight.sel(_2).mean("n") else: # `node` without `ref_nodes` # Arbitrary value k = 1 / 3.0 # As k tends to 1, converge towards equal weights target = xr.DataArray(1.0 / len(t_modes)) # Scale weights in convergence year # - As k tends to 0, converge towards the same node's base shares. weight.loc[_1] = k * target + (1 - k) * weight.sel(n=node, **y0) # Interpolate linearly between y0 and yC # NB this will not work if yC is before the final period; it will leave NaN after yC weight = weight.interpolate_na(dim="y") return genno.Quantity(weight)
[docs]def smooth(qty: "AnyQuantity") -> "AnyQuantity": """Smooth `qty` (e.g. PRICE_COMMODITY) in the ``y`` dimension.""" from genno.operator import add, concat # General smoothing result = add(0.25 * qty.shift(y=-1), 0.5 * qty, 0.25 * qty.shift(y=1)) y = qty.coords["y"].values # Shorthand for weights def _w(values, years): return genno.Quantity(values, coords={"y": years}, units="") # First period r0 = (qty * _w([0.4, 0.4, 0.2], y[:3])).sum("y").expand_dims(dict(y=y[:1])) # Final period. “closer to the trend line” # NB the inherited R file used a formula equivalent to weights like [-⅛, 0, ⅜, ¾]; # didn't make much sense. r_m1 = (qty * _w([0.2, 0.2, 0.6], y[-3:])).sum("y").expand_dims(dict(y=y[-1:])) # apply_units() is to work around khaeru/genno#64 # TODO remove when fixed upstream return apply_units(concat(r0, result.sel(y=y[1:-1]), r_m1), qty.units)
def _add_transport_data(func, c: "Computer", name: str, *, key) -> None: """Add data from `key` to the target scenario. Adds one task to `c` that uses :func:`.add_par_data` to store the data from `key` on "scenario". Also updates the "add transport data" computation by appending the new task. """ c.add(f"add {name}", "add_par_data", "scenario", key, "dry_run", strict=True) c.graph["add transport data"].append(f"add {name}") @Operator.define(helper=_add_transport_data) def transport_data(*args): """No action. This exists to connect :func:`._add_transport_data` to :meth:`genno.Computer.add`. """ pass # pragma: no cover
[docs]def transport_check(scenario: "Scenario", ACT: "AnyQuantity") -> pd.Series: """Reporting operator for :func:`.check`.""" info = ScenarioInfo(scenario) # Mapping from check name → bool checks = {} # Correct number of outputs ACT_lf = ACT.sel(t=["transport freight load factor", "transport pax load factor"]) checks["'transport * load factor' technologies are active"] = len( ACT_lf ) == 2 * len(info.Y) * (len(info.N) - 1) # # Force the check to fail # checks['(fail for debugging)'] = False return pd.Series(checks)
[docs]def votm(gdp_ppp_cap: "AnyQuantity") -> "AnyQuantity": """Calculate value of time multiplier. A value of 1 means the VoT is equal to the wage rate per hour. Parameters ---------- gdp_ppp_cap PPP GDP per capita. """ from genno.operator import assign_units u = gdp_ppp_cap.units assert_units(gdp_ppp_cap, "kUSD / passenger / year") n = gdp_ppp_cap.coords["n"].data result = 1 / ( 1 + assign_units( np.exp( (genno.Quantity(30, units=u).expand_dims({"n": n}) - gdp_ppp_cap) / 20 ), units="", ) ) assert_units(result, "") return result