EntsoeSource

Downloads electricity data from the ENTSOE Transparency Platform: load, day-ahead prices, generation, cross-border transfer capacity and installed capacity.

Setup

You need an ENTSOE API key:

1. Register at ENTSOE Transparency Platform
2. Request an API key from your account settings

Basic Usage

from epftoolbox2.data.sources import EntsoeSource
import os
import pandas as pd

source = EntsoeSource(
    country_code="PL",
    api_key=os.environ.get("ENTSOE_API_KEY"),
    type=["load", "price"],
    prefer_15min=False,
)

df = source.fetch(
    start=pd.Timestamp("2024-01-01", tz="UTC"),
    end=pd.Timestamp("2024-06-01", tz="UTC"),
)

Parameters

Parameter	Type	Required	Description
country_code	str	Yes	Area code (see table below). Prices need a bidding zone, e.g. "DE_LU" rather than "DE".
api_key	str	Yes	ENTSOE API key. It is sent in the SECURITY_TOKEN request header, never in the URL, and error messages never contain it.
type	List[str]	Yes	Data types: load, price, generation, transfer_capacity, installed_capacity
prefer_15min	bool	No	Defaults to False. Selects the grid of the price column: False gives hourly prices, True gives 15-minute prices. See Price resolution.
neighbours	List[str] or str	With transfer_capacity	Defaults to None. Neighbouring bidding zones for transfer_capacity, e.g. ["DE_LU", "CZ"] or "DE_LU,CZ". Required with that type and rejected without it.
prefix	str	No	Defaults to "". Prepended to every column as f"{prefix}_{column}", so several zones can be combined in one pipeline.

Returned range

fetch(start, end) returns the half-open range [start, end) with a UTC index. Data is requested in 3-month chunks. A chunk that keeps failing raises an error instead of leaving a gap.

Daily values (load_forecast_daily_min/max) are stamped at local midnight of the day they describe: the market’s local midnight when this source is part of a DataPipeline, and this area’s own otherwise (see Week-ahead daily values and neighbouring areas below). When that midnight lies before start (for example a Warsaw day starts at 23:00 UTC the evening before), the value that covers start is reported at start, so the first day is not lost.

---

Data Types

Load Data (type=["load"])

Column	Document	Description
load_actual	A65 / A16	Actual total load (MW)
load_forecast	A65 / A01	Day-ahead total load forecast (MW)
load_forecast_daily_min	A65 / A31	Week-ahead forecast of the day’s minimum load (MW), one value per local day
load_forecast_daily_max	A65 / A31	Week-ahead forecast of the day’s maximum load (MW), one value per local day

Load keeps the resolution ENTSO-E publishes (15 minutes for PL and DE_LU in 2024–2025). The daily columns hold one value at local midnight and are empty in between. A ResampleTransformer(freq="1h") after TimezoneTransformer spreads each value over its own local day.

Week-ahead daily values and neighbouring areas

load_forecast_daily_min/max (A31) holds one value per this area’s own local calendar day D. Combining this source with another area’s in one market pipeline needs that value on the market’s local day D, not necessarily this area’s: a Finnish (EET) neighbour’s day starts an hour before a Polish (CET) market’s, so without correcting for that, its daily value would land an hour into the wrong side of midnight, and around a DST switch even on the wrong calendar day.

EntsoeSource fixes this automatically when it is part of a DataPipeline: it stamps the value at the market’s local midnight of its own day D, instead of this area’s own midnight, so the row already sits where TimezoneTransformer’s default snap (anchored to the market zone) expects it, a no-op for this column. Used standalone, or when this area already is the market’s own, the value is stamped at this area’s own midnight.

# FI's week-ahead daily min/max land on Warsaw's own local days, not Helsinki's
pipeline = (
    DataPipeline(timezone="Europe/Warsaw")
    .add_source(EntsoeSource("PL", api_key=api_key, type=["load"]))
    .add_source(EntsoeSource("FI", api_key=api_key, type=["load"], prefix="fi"))
    .add_transformer(TimezoneTransformer())
)

TimezoneTransformer’s default daily_columns recognise a prefixed neighbour’s column (fi_load_forecast_daily_min) automatically too. See TimezoneTransformer.

Price Data (type=["price"])

Column	Document	Description
price	A44	Day-ahead price of the single day-ahead coupling (SDAC) in the bidding zone (EUR/MWh)

Price resolution

The day-ahead market moved from 60-minute to 15-minute products on 2025-10-01. Before that date most zones publish hourly prices, after it 15-minute prices. EntsoeSource keeps every period and returns one uniform grid:

prefer_15min	Grid	Hourly periods	15-minute periods
False (default)	1 hour	as published	averaged into the hour (an hour with a missing quarter-hour is left empty)
True	15 minutes	step-filled: each quarter-hour gets its hour’s price	as published

Where a zone publishes the same interval at two resolutions, the one matching the grid is used. 30-minute periods are treated the same way.

Caution

Only the SDAC price is returned. For DE_LU and AT the API also publishes a second, 15-minute series (classificationSequence 2, a separate auction) on the same timestamps. It differs from the coupling price by up to several hundred EUR/MWh and is ignored. Before 2025-10-01, prefer_15min=True therefore gives the hourly SDAC price step-filled to 15 minutes, not the 15-minute auction.

Caution

Bidding zones: prices are published per bidding zone. For Germany use "DE_LU" (since 2018-10-01, "DE_AT_LU" before). Codes that are not bidding zones ("DE", "DK", "NO", "SE", "IT", "LU", the German control areas, …) raise a ValueError naming the zone to use when type includes "price". They are accepted for load and generation.

Generation Data (type=["generation"])

Actual generation per production type (A75) as generation_<type>, and the day-ahead wind and solar forecast (A69) as generation_<type>_forecast. Where a production type also consumes (pumped-storage pumping, and for some zones every type, mostly zeros), the consumption is a separate column generation_<type>_consumption. It is never mixed into the generation column.

Column	PsrType	Description
generation_biomass	B01	Biomass (MW)
generation_fossil_brown_coal_lignite	B02	Lignite (MW)
generation_fossil_coal_derived_gas	B03	Coal-derived gas (MW)
generation_fossil_gas	B04	Natural gas (MW)
generation_fossil_hard_coal	B05	Hard coal (MW)
generation_fossil_oil	B06	Oil (MW)
generation_fossil_oil_shale	B07	Oil shale (MW)
generation_fossil_peat	B08	Peat (MW)
generation_geothermal	B09	Geothermal (MW)
generation_hydro_pumped_storage	B10	Pumped-storage hydro, generating (MW)
generation_hydro_run_of_river_and_pondage	B11	Run-of-river hydro (MW)
generation_hydro_water_reservoir	B12	Reservoir hydro (MW)
generation_marine	B13	Marine/tidal (MW)
generation_nuclear	B14	Nuclear (MW)
generation_other_renewable	B15	Other renewable (MW)
generation_solar	B16	Solar (MW)
generation_waste	B17	Waste (MW)
generation_wind_offshore	B18	Offshore wind (MW)
generation_wind_onshore	B19	Onshore wind (MW)
generation_other	B20	Other (MW)
generation_energy_storage	B25	Energy storage (MW)
generation_<type>_consumption	any	Consumption reported for that type, e.g. generation_hydro_pumped_storage_consumption (MW)
generation_solar_forecast, generation_wind_onshore_forecast, generation_wind_offshore_forecast	B16, B19, B18	Day-ahead forecast (MW)

Column names are the snake_case of ENTSO-E’s production type names. An unknown code becomes generation_<code>, e.g. generation_b99.

Transfer Capacity (type=["transfer_capacity"])

Day-ahead forecasted transfer capacity (A61, contract type daily) in MW, for each zone in neighbours:

Column	Description
transfer_capacity_export_<neighbour>	Capacity from country_code to the neighbour, e.g. transfer_capacity_export_es
transfer_capacity_import_<neighbour>	Capacity from the neighbour to country_code

source = EntsoeSource("FR", api_key=key, type=["transfer_capacity"], neighbours=["ES", "CH"])

Note

Many borders publish no day-ahead forecasted capacity, in particular borders inside a flow-based capacity calculation region (Core since 2022-06-08). The API then answers “No matching data”: the column is missing and a warning is logged. In a check for 2024-10-01..02, PL→DE_LU, PL→LT, LT→PL and SE_4→PL returned nothing (PL→DE_LU also nothing in March 2021), while FR→ES and CH→DE_LU returned hourly values.

Installed Capacity (type=["installed_capacity"])

Installed generation capacity per production type (A68, year-ahead) in MW, as installed_capacity_<type> with the same type names as generation (e.g. installed_capacity_solar, installed_capacity_wind_onshore). There is one value per year, stamped at local midnight of 1 January. If that lies before start, the value is reported at start. A ResampleTransformer(freq="1h") after TimezoneTransformer carries each value forward over its year.

Combining zones with prefix

from epftoolbox2.pipelines import DataPipeline

pipeline = (
    DataPipeline()
    .add_source(EntsoeSource("PL", api_key=key, type=["load", "price"]))
    .add_source(EntsoeSource("CZ", api_key=key, type=["price"], prefix="cz"))            # cz_price
    .add_source(EntsoeSource("DE_LU", api_key=key, type=["price"], prefix="de"))         # de_price
)

Not offered: unavailability of generation units (A77/A80)

Outage documents are versioned, but the API returns only the latest revision of each outage by default. With mRID it returns every revision, but their createdDateTime is not the original publication time: for Polish outages active on 2024-10-01..02, all 75 latest revisions, and all three revisions of the one outage checked, were dated October 2025. So the values known at issue time cannot be rebuilt from the API (it would also take one request per outage), and the latest revision is look-ahead. These documents are left out.

---

Information available at forecast time

A day-ahead forecast for delivery day D+1 is issued on day D before the market’s gate closure at 12:00 CET. Commission Regulation (EU) No 543/2013 sets when each series must be published. Lags are relative to the row’s local date D, as produced by LagTransformer (x_d-1 is yesterday, x_d+1 tomorrow):

Column	Publication deadline (543/2013)	Safe as a predictor for a forecast issued on D
price	No later than 1 hour after gate closure (Art. 12)	price (delivery D, cleared on D-1) and _d-k. price_d+1 is the target
load_actual	No later than 1 hour after the operating period (Art. 6)	_d-1 and older only. Unlagged load_actual is look-ahead for the hours after the issue time
load_forecast	2 hours before gate closure (10:00 CET) on the day before delivery, updated when significant changes occur (Art. 6)	load_forecast and _d+1 if the forecast is issued at or after 10:00. _d+2 and later are published on D+1 or later: look-ahead
load_forecast_daily_min/max	Fridays, 2 hours before gate closure, for Monday–Sunday of the following week. Updated when significant changes occur (Art. 6)	_d+0 to _d+2 always. _d+3 when issued at or after 10:00 (on Friday mornings only the week up to Sunday is known). Larger horizons leak on some weekdays
generation_* (actual, incl. _consumption)	No later than 1 hour after the operational period (Art. 16)	_d-1 and older only
generation_*_forecast	18:00 Brussels time on the day before delivery, updated during intraday trading with at least one update at 08:00 on the delivery day (Art. 14)	_d+0 and older. _d+1 is published after gate closure: not guaranteed at issue time
transfer_capacity_*	”Sufficiently in advance of the allocation process” (Art. 11): before the day-ahead gate closure, no fixed hour	_d+0 and older. _d+1 is known before 12:00 on D but at no guaranteed hour, so borderline for an early issue time
installed_capacity_*	Annually, no later than one week before the end of the previous year (Art. 14)	The current year’s value (and _d+h within the same year). The API serves the latest revision

The API serves the latest revision of each series, not the version published at the deadline, so forecasts that are revised after the issue time (load forecasts, wind and solar forecasts) can still carry a small amount of look-ahead even within these limits.

---

Errors and retries

• “No matching data” for a chunk is logged as a warning and the chunk is skipped (for example future dates or a document type the area does not publish). A warning is also logged when the data starts more than a day after start (e.g. DE_LU before 2018-10-01).
• HTTP 429 and 5xx responses and network errors are retried up to 5 times with exponential backoff, honouring the Retry-After header. Other HTTP errors raise EntsoeApiError with ENTSO-E’s reason text, e.g. HTTP 401: Authentication failed.

---

Supported Countries & Area Codes

Any key of epftoolbox2.data.sources.entsoe.AREAS is accepted. Common bidding zones:

Code	Area	Area Code
AT	Austria	10YAT-APG------L
BE	Belgium	10YBE----------2
BG	Bulgaria	10YCA-BULGARIA-R
CH	Switzerland	10YCH-SWISSGRIDZ
CZ	Czechia	10YCZ-CEPS-----N
DE_LU	Germany-Luxembourg	10Y1001A1001A82H
DK_1	Denmark West	10YDK-1--------W
DK_2	Denmark East	10YDK-2--------M
EE	Estonia	10Y1001A1001A39I
ES	Spain	10YES-REE------0
FI	Finland	10YFI-1--------U
FR	France	10YFR-RTE------C
GB	Great Britain	10YGB----------A
GR	Greece	10YGR-HTSO-----Y
HR	Croatia	10YHR-HEP------M
HU	Hungary	10YHU-MAVIR----U
IE_SEM	Ireland (SEM)	10Y1001A1001A59C
IT_NORD	Italy North	10Y1001A1001A73I
LT	Lithuania	10YLT-1001A0008Q
LV	Latvia	10YLV-1001A00074
NL	Netherlands	10YNL----------L
NO_1 … NO_5	Norway zones	10YNO-1--------2 …
PL	Poland	10YPL-AREA-----S
PT	Portugal	10YPT-REN------W
RO	Romania	10YRO-TEL------P
RS	Serbia	10YCS-SERBIATSOV
SE_1 … SE_4	Sweden zones	10Y1001A1001A44P …
SI	Slovenia	10YSI-ELES-----O
SK	Slovakia	10YSK-SEPS-----K

Country and control-area codes such as DE (10Y1001A1001A83F), DK, NO, SE, IT, LU and IE work for load and generation but have no day-ahead price.

Tip

Use the code (e.g., "PL") in your source configuration. The area code is used internally for API requests.

---

Caching

The cache key contains the area, the types, prefer_15min and a schema version of the parser. Changing any of them uses a new cache directory under .cache/sources/. Directories for other values are not read and can be deleted.

---

Output

Data is returned with a UTC DatetimeIndex:

print(df.head())
#                            load_actual  price
# 2024-01-01 00:00:00+00:00     12508.71   5.75
# 2024-01-01 01:00:00+00:00     11845.20   5.27

Load and generation have one row per published interval (often 15 minutes), so an hourly price column is empty on the :15, :30 and :45 rows until you resample, e.g. with ResampleTransformer(freq="1h").