pypsa-eur/scripts/build_load_data.py

# -*- coding: utf-8 -*-
# SPDX-FileCopyrightText: : 2020 @JanFrederickUnnewehr, The PyPSA-Eur Authors
#
# SPDX-License-Identifier: MIT

"""
This rule downloads the load data from `Open Power System Data Time series.

<https://data.open-power-system-data.org/time_series/>`_. For all countries in
the network, the per country load timeseries with suffix
``_load_actual_entsoe_transparency`` are extracted from the dataset. After
filling small gaps linearly and large gaps by copying time-slice of a given
period, the load data is exported to a ``.csv`` file.

Relevant Settings
-----------------

.. code:: yaml

    snapshots:

    load:
        interpolate_limit:
        time_shift_for_large_gaps:
        manual_adjustments:


.. seealso::
    Documentation of the configuration file ``config.yaml`` at
    :ref:`load_cf`

Inputs
------

- ``data/load_raw.csv``:

Outputs
-------

- ``resources/load.csv``:
"""

import logging

logger = logging.getLogger(__name__)
import dateutil
import numpy as np
import pandas as pd
from _helpers import configure_logging
from pandas import Timedelta as Delta


def load_timeseries(fn, years, countries, powerstatistics=True):
    """
    Read load data from OPSD time-series package version 2020-10-06.

    Parameters
    ----------
    years : None or slice()
        Years for which to read load data (defaults to
        slice("2018","2019"))
    fn : str
        File name or url location (file format .csv)
    countries : listlike
        Countries for which to read load data.
    powerstatistics: bool
        Whether the electricity consumption data of the ENTSOE power
        statistics (if true) or of the ENTSOE transparency map (if false)
        should be parsed.

    Returns
    -------
    load : pd.DataFrame
        Load time-series with UTC timestamps x ISO-2 countries
    """
    logger.info(f"Retrieving load data from '{fn}'.")

    pattern = "power_statistics" if powerstatistics else "transparency"
    pattern = f"_load_actual_entsoe_{pattern}"

    def rename(s):
        return s[: -len(pattern)]

    def date_parser(x):
        return dateutil.parser.parse(x, ignoretz=True)

    return (
        pd.read_csv(fn, index_col=0, parse_dates=[0], date_parser=date_parser)
        .filter(like=pattern)
        .rename(columns=rename)
        .dropna(how="all", axis=0)
        .rename(columns={"GB_UKM": "GB"})
        .filter(items=countries)
        .loc[years]
    )


def consecutive_nans(ds):
    return (
        ds.isnull()
        .astype(int)
        .groupby(ds.notnull().astype(int).cumsum()[ds.isnull()])
        .transform("sum")
        .fillna(0)
    )


def fill_large_gaps(ds, shift):
    """
    Fill up large gaps with load data from the previous week.

    This function fills gaps ragning from 3 to 168 hours (one week).
    """
    shift = Delta(shift)
    nhours = shift / np.timedelta64(1, "h")
    if (consecutive_nans(ds) > nhours).any():
        logger.warning(
            "There exist gaps larger then the time shift used for "
            "copying time slices."
        )
    time_shift = pd.Series(ds.values, ds.index + shift)
    return ds.where(ds.notnull(), time_shift.reindex_like(ds))


def nan_statistics(df):
    def max_consecutive_nans(ds):
        return (
            ds.isnull()
            .astype(int)
            .groupby(ds.notnull().astype(int).cumsum())
            .sum()
            .max()
        )

    consecutive = df.apply(max_consecutive_nans)
    total = df.isnull().sum()
    max_total_per_month = df.isnull().resample("m").sum().max()
    return pd.concat(
        [total, consecutive, max_total_per_month],
        keys=["total", "consecutive", "max_total_per_month"],
        axis=1,
    )


def copy_timeslice(load, cntry, start, stop, delta, fn_load=None):
    start = pd.Timestamp(start)
    stop = pd.Timestamp(stop)
    if start in load.index and stop in load.index:
        if start - delta in load.index and stop - delta in load.index and cntry in load:
            load.loc[start:stop, cntry] = load.loc[
                start - delta : stop - delta, cntry
            ].values
        elif fn_load is not None:
            duration = pd.date_range(freq="h", start=start - delta, end=stop - delta)
            load_raw = load_timeseries(fn_load, duration, [cntry], powerstatistics)
            load.loc[start:stop, cntry] = load_raw.loc[
                start - delta : stop - delta, cntry
            ].values


def manual_adjustment(load, fn_load, powerstatistics):
    """
    Adjust gaps manual for load data from OPSD time-series package.

     1. For the ENTSOE power statistics load data (if powerstatistics is True)

     Kosovo (KV) and Albania (AL) do not exist in the data set. Kosovo gets the
     same load curve as Serbia and Albania the same as Macdedonia, both scaled
     by the corresponding ratio of total energy consumptions reported by
     IEA Data browser [0] for the year 2013.

     2. For the ENTSOE transparency load data (if powerstatistics is False)

     Albania (AL) and Macedonia (MK) do not exist in the data set. Both get the
     same load curve as Montenegro,  scaled by the corresponding ratio of total energy
     consumptions reported by  IEA Data browser [0] for the year 2016.

     [0] https://www.iea.org/data-and-statistics?country=WORLD&fuel=Electricity%20and%20heat&indicator=TotElecCons


     Parameters
     ----------
     load : pd.DataFrame
         Load time-series with UTC timestamps x ISO-2 countries
     powerstatistics: bool
         Whether argument load comprises the electricity consumption data of
         the ENTSOE power statistics or of the ENTSOE transparency map
    load_fn: str
         File name or url location (file format .csv)

     Returns
     -------
     load : pd.DataFrame
         Manual adjusted and interpolated load time-series with UTC
         timestamps x ISO-2 countries
    """
    if powerstatistics:
        if "MK" in load.columns:
            if "AL" not in load.columns or load.AL.isnull().values.all():
                load["AL"] = load["MK"] * (4.1 / 7.4)
        if "RS" in load.columns:
            if "KV" not in load.columns or load.KV.isnull().values.all():
                load["KV"] = load["RS"] * (4.8 / 27.0)

        copy_timeslice(
            load, "GR", "2015-08-11 21:00", "2015-08-15 20:00", Delta(weeks=1)
        )
        copy_timeslice(
            load, "AT", "2018-12-31 22:00", "2019-01-01 22:00", Delta(days=2)
        )
        copy_timeslice(
            load, "CH", "2010-01-19 07:00", "2010-01-19 22:00", Delta(days=1)
        )
        copy_timeslice(
            load, "CH", "2010-03-28 00:00", "2010-03-28 21:00", Delta(days=1)
        )
        # is a WE, so take WE before
        copy_timeslice(
            load, "CH", "2010-10-08 13:00", "2010-10-10 21:00", Delta(weeks=1)
        )
        copy_timeslice(
            load, "CH", "2010-11-04 04:00", "2010-11-04 22:00", Delta(days=1)
        )
        copy_timeslice(
            load, "NO", "2010-12-09 11:00", "2010-12-09 18:00", Delta(days=1)
        )
        # whole january missing
        copy_timeslice(
            load,
            "GB",
            "2010-01-01 00:00",
            "2010-01-31 23:00",
            Delta(days=-365),
            fn_load,
        )
        # 1.1. at midnight gets special treatment
        copy_timeslice(
            load,
            "IE",
            "2016-01-01 00:00",
            "2016-01-01 01:00",
            Delta(days=-366),
            fn_load,
        )
        copy_timeslice(
            load,
            "PT",
            "2016-01-01 00:00",
            "2016-01-01 01:00",
            Delta(days=-366),
            fn_load,
        )
        copy_timeslice(
            load,
            "GB",
            "2016-01-01 00:00",
            "2016-01-01 01:00",
            Delta(days=-366),
            fn_load,
        )

    else:
        if "ME" in load:
            if "AL" not in load and "AL" in countries:
                load["AL"] = load.ME * (5.7 / 2.9)
            if "MK" not in load and "MK" in countries:
                load["MK"] = load.ME * (6.7 / 2.9)
        copy_timeslice(
            load, "BG", "2018-10-27 21:00", "2018-10-28 22:00", Delta(weeks=1)
        )

    return load


if __name__ == "__main__":
    if "snakemake" not in globals():
        from _helpers import mock_snakemake

        snakemake = mock_snakemake("build_load_data")

    configure_logging(snakemake)

    powerstatistics = snakemake.config["load"]["power_statistics"]
    interpolate_limit = snakemake.config["load"]["interpolate_limit"]
    countries = snakemake.config["countries"]
    snapshots = pd.date_range(freq="h", **snakemake.config["snapshots"])
    years = slice(snapshots[0], snapshots[-1])
    time_shift = snakemake.config["load"]["time_shift_for_large_gaps"]

    load = load_timeseries(snakemake.input[0], years, countries, powerstatistics)

    if snakemake.config["load"]["manual_adjustments"]:
        load = manual_adjustment(load, snakemake.input[0], powerstatistics)

    logger.info(f"Linearly interpolate gaps of size {interpolate_limit} and less.")
    load = load.interpolate(method="linear", limit=interpolate_limit)

    logger.info(
        "Filling larger gaps by copying time-slices of period " f"'{time_shift}'."
    )
    load = load.apply(fill_large_gaps, shift=time_shift)

    assert not load.isna().any().any(), (
        "Load data contains nans. Adjust the parameters "
        "`time_shift_for_large_gaps` or modify the `manual_adjustment` function "
        "for implementing the needed load data modifications."
    )

    load.to_csv(snakemake.output[0])