enedis-with-holidays (TsFile)

This is the Apache TsFile conversion of theforecastingcompany/enedis-with-holidays.

Five years (2020-07-18 → 2025-07-17) of the French total electricity consumption (consommation_totale) published by Enedis, at three frequencies: 30min, 6h, and D (daily). Each row carries the matching covariates (smoothed observed temperature, smoothed climatological-normal temperature, French public holiday flag) and nine forecast windows anchored on three 2024 holidays for backtesting.

The dataset is designed to evaluate how well forecasting models capture holiday effects. French public holidays cause large, predictable drops in electricity consumption that are notoriously hard to forecast with seasonality alone — Labour Day (May 1), Armistice (Nov 11), and the Christmas / New-Year cluster are all included explicitly, with multiple forecast-creation dates per holiday so you can compare model behaviour across short, medium, and long holiday lead times. The is_france_holiday future covariate is the hook for testing covariate-aware models against ones that have to infer the holiday calendar themselves.

Files

.
├── README.md
└── data/
    └── enedis-with-holidays.tsfile   (1,608,535 bytes)

TsFile layout

The original Hugging Face dataset stores each frequency as a single row whose target / covariates are nested arrays (GluonTS / Chronos style, 3 rows total). For TsFile this is flattened into one table with three devices (TsFile tag = item_id), each device being one frequency with its own time axis.

• Table: enedis_with_holidays
• TAG (device dimension): item_id — one of enedis_bilan_30min, enedis_bilan_6h, enedis_bilan_D
• Time: synthesized from the row's start + freq, stored as INT64 epoch milliseconds
• FIELD columns (all FLOAT): the target plus three covariate channels, expanded along the time axis

device (item_id)	freq	rows (T)	time span
enedis_bilan_30min	30min	87 648	2020-07-18 → 2025-07-17
enedis_bilan_6h	6h	7 305	2020-07-18 → 2025-07-17
enedis_bilan_D	D	1 827	2020-07-18 → 2025-07-17

Total: 96 780 rows across the three devices.

Columns

column	category	type	meaning
Time	TIME	INT64	epoch ms, synthesized from start + freq
item_id	TAG	STRING	device id, encodes the frequency
consommation_totale	FIELD	FLOAT	total French electricity consumption, energy delivered per bucket in Wh
temperature_reelle_lissee	FIELD	FLOAT	smoothed observed temperature (°C) — history-only covariate
is_france_holiday	FIELD	FLOAT	binary 0/1, 1 on French public holidays — known-future covariate
temperature_normale_lissee	FIELD	FLOAT	smoothed climatological-normal temperature (°C) — known-future covariate

Target & covariates

• consommation_totale — total French electricity consumption, expressed as energy delivered during the bucket in Wh. At daily granularity the value is the total daily energy (≈ 1.2–1.5 TWh per day for France). The original float32 values are kept as-is (no unit conversion).
• temperature_reelle_lissee (history-only) — smoothed observed temperature (°C).
• is_france_holiday (known-future) — binary, 1 on French public holidays.
• temperature_normale_lissee (known-future) — smoothed climatological-normal temperature (°C).

For the 6h and D devices the consumption channel is the sum over the bucket (units stay Wh of delivered energy per bucket); the holiday flag is the max (any holiday inside marks the whole bucket); temperatures are bucket means. The post-2024-10-04 15-min era in the raw Enedis data is aggregated to 30-min mean before the half-hourly device is emitted, so the timeline is uniform across the five-year span.

Forecast windows

The original dataset ships nine backtest windows = three holidays × three forecast-creation dates each. These are per-series index metadata, not time-series points, so they are not stored inside the TsFile. They are preserved verbatim in this repo for reference:

holiday	date	FCDs (days before)	horizons (days)
Labour Day	2024-05-01	20, 10, 5	20, 10, 5
Armistice	2024-11-11	20, 10, 5	20, 10, 5
Christmas / New Year	2024-12-25	28, 18, 13	28, 18, 13

In the source data the FCD indexes (window_fcd_idxs) and horizons (window_horizons) are expressed in steps of each device's own frequency (the 30min device's horizons are in half-hours, the 6h device's in six-hour buckets, etc.). See the original dataset for the exact integer arrays and the ground-truth slicing convention.

Reading the TsFile

from tsfile import TsFileReader

reader = TsFileReader("data/enedis-with-holidays.tsfile")

# inspect tables / columns
for name, table in reader.get_all_table_schemas().items():
    print(name, [(c.get_column_name(), c.get_data_type()) for c in table.get_columns()])

# query all field/tag columns of the table
cols = ["item_id", "consommation_totale", "temperature_reelle_lissee",
        "is_france_holiday", "temperature_normale_lissee"]
with reader.query_table("enedis_with_holidays", cols, batch_size=65536) as rs:
    while (batch := rs.read_arrow_batch()) is not None:
        df = batch.to_pandas()   # Time column is added automatically
        ...

Notes on conversion

Columns from the source dataset that are not carried into the time-series data (constant or already encoded elsewhere):

• freq — encoded into the Time axis spacing and the item_id suffix.
• source, source_item_id — provenance constants (see Source below).
• target_names, past_feat_dynamic_real_names, feat_dynamic_real_names — fixed channel-name constants, documented in the column table above.
• window_fcd_idxs, window_horizons — backtest window metadata, kept for reference (see Forecast windows) rather than as time-series points.

Source & license

Built from the bilan électrique demi-heure series published by Enedis Open Data and mirrored on data.gouv.fr. The covariates are publicly available French public-holiday calendars and Météo-France smoothed temperature references. Redistributed under CC-BY-4.0 with credit to Enedis as the upstream data provider.

• Original Hugging Face dataset: theforecastingcompany/enedis-with-holidays
• Upstream data provider: Enedis Open Data

Citation

If you use this dataset, please credit Enedis as the upstream data provider and link back to the original Hugging Face repository.