padelf

Easy Pandas DataFrame-Access to publicly available electric load forecasting datasets

padelf provides a minimal Python API to download, cache, and standardize electric load forecasting datasets for research. Every dataset is returned as a pandas DataFrame with a UTC DateTimeIndex and a standardized consumption_kW column.

Installation

pip install padelf

Quick Start

import padelf

# See what's available
padelf.list_datasets()
# ['AEMO', 'ELD', 'ENTSO-E', 'GEFCOM12', 'IHPC', 'ISO-NE', 'NYISO', 'OPSD', 'Pecan-Street', 'RTE-France', 'VEA']

# Load a dataset — one line, sensible defaults
df = padelf.get_dataset("OPSD")
print(df.head())

Output:

                                                     consumption_kW  DE_solar_generation_actual  DE_wind_onshore_generation_actual
datetime
2015-01-01 00:00:00+00:00       41209.0                        NaN                            7568.0
2015-01-01 01:00:00+00:00       40029.0                        NaN                            7666.0
2015-01-01 02:00:00+00:00       38891.0                        NaN                            7637.0

What You Get

Every call to get_dataset() returns a DataFrame with:

• DateTimeIndex — UTC timezone, equidistant at the dataset's native resolution
• consumption_kW — Load/consumption column, unit-converted to kilowatts
• Additional columns — As available in the original dataset (e.g., temperature, solar generation)

Optional Parameters

df = padelf.get_dataset(
        "OPSD",
        resolution="15min",       # Resample to 15-minute intervals
        consumption_unit="MW",    # Keep original MW units
        interpolate_limit="4h",   # Fill gaps up to 4 hours
        cache_dir="/tmp/padelf",  # Custom cache location
)

Available Datasets

Dataset	Abbreviation	Resolution	Region	Status
Open Power System Data	OPSD	60 min	Europe	✅ Ready
Individual Household Power Consumption	IHPC	1 min	France	✅ Ready
ElectricityLoadDiagrams20112014	ELD	15 min	Portugal	✅ Ready
5359 industrial VEA load profiles	VEA	15 min	Germany	✅ Ready
GEFCom 2012	GEFCOM12	60 min	US	⏩ Planned
ENTSO-E Transparency	ENTSO-E	60 min	Europe	⏩ Planned
ISO New England	ISO-NE	60 min	US	⏩ Planned
NYISO	NYISO	5 min	US	⏩ Planned
AEMO	AEMO	60 min	Australia	⏩ Planned
RTE France	RTE-France	30 min	France	⏩ Planned
Pecan Street	Pecan Street	15 min	US	⏩ Planned

Repository Structure

The project uses a src layout with per-dataset YAML configs:

.
├── README.md
├── pyproject.toml                # Build config (hatchling backend)
├── mkdocs.yml                    # Documentation site config
├── LICENSE
├── src/padelf/
│   ├── __init__.py               # Public API: list_datasets(), get_dataset()
│   ├── loader.py                 # Core loader logic: download, cache, parse, standardize
│   ├── utils.py                  # Unit conversion, gap interpolation, resampling
│   └── configs/
│       ├── _template.yaml        # Template for new loader configs
│       ├── OPSD.yaml             # Ready
│       ├── IHPC.yaml             # Ready
│       ├── ELD.yaml              # Ready
│       ├── VEA.yaml              # Ready
│       ├── GEFCOM12.yaml         # Ready (source URL intermittent)
│       ├── ENTSO-E.yaml          # API placeholder
│       ├── ISO-NE.yaml           # API placeholder
│       ├── NYISO.yaml            # API placeholder
│       ├── AEMO.yaml             # API placeholder
│       ├── RTE-France.yaml       # API placeholder
│       └── Pecan-Street.yaml     # API placeholder
├── docs/                         # mkdocs source files
│   ├── index.md
│   ├── getting-started.md
│   ├── api.md
│   └── datasets.md
└── tests/
    ├── test_loader.py
    ├── test_utils.py
    └── test_smoke.py

How It Works

The loader architecture follows a per-dataset config pattern. Each YAML file in src/padelf/configs/ defines a dataset's download URL, file format, column mappings, unit, and preprocessing parameters. When get_dataset() is called, loader.py reads the corresponding config, downloads the file (or uses a local cache), parses it, and applies standardization via utils.py: the load column is renamed to consumption_kW with automatic unit conversion (MW, kWh, MWh to kW), the index is converted to an equidistant UTC DateTimeIndex, gaps up to 2 hours are interpolated by default, and optional resampling is applied if requested. Datasets flagged with requires_api: true in their config raise NotImplementedError with a descriptive message -- these are placeholders for future implementation.

Adding a New Loader

See the Loader Developer Guide for details on the loader architecture and how to add new datasets.

API Placeholder Pattern

Six datasets (ENTSO-E, ISO-NE, NYISO, AEMO, RTE-France, Pecan-Street) are currently configured as API placeholders. Their YAML configs exist with requires_api: true, and calling get_dataset() on them raises NotImplementedError. To convert a placeholder into a working loader, remove the requires_api flag and either provide a direct download URL or implement API-specific download logic in loader.py. Note that ENTSO-E and ISO-NE have direct CSV downloads available and could be implemented as file-based loaders without API integration.

Original Catalog

To explore more datasets, check out the original PADELF Repository.

Citation

If this work has helped you with your scientific work, we would appreciate a proper mention. ❤️

@inproceedings{baur2024datasets,
    title     = {Publicly Available Datasets For Electric Load Forecasting -- An Overview},
    author    = {Baur, Lukas and Chandramouli, Vignesh and Sauer, Alexander},
    booktitle = {6th Conference on Production Systems and Logistics (CPSL 2024)},
    year      = {2024},
    doi       = {10.15488/17659}
}

Changelog

See the Changelog File

License

MIT