OpenSTEF - LF Energy

Back To All Projects

Github

Mailing List

Python Package

Website

Wiki

Roadmap

Dashboard Docs

Documentation

View Next Project

OpenSTEF: Short Term Energy Forecasting

OpenSTEF is an open source Python package that provides a comprehensive portfolio of machine learning pipelines to create accurate short-term energy forecasts. It is collaboratively developed by a growing community of grid operators, technology vendors, researchers, and energy-sector experts, and is hosted at LF Energy. It generates probabilistic forecasts for hours to days ahead for any given energy signal, enabling grid operators, energy companies, and researchers to anticipate congestion, support grid safety analysis, and optimize flexible assets.

Why OpenSTEF?

The energy transition is causing both a rapid growth in net grid loads and fundamental changes in the patterns of power supply and distribution. Mitigating grid congestion and optimizing the use of assets has become the number one mission critical challenge for utilities who are managing this change across existing grid networks. As electricity consumption and generation become increasingly influenced by weather conditions and market dynamics, load patterns on the grid are growing more complex. Anticipating these patterns is therefore central to present day grid operations, creating a pressing need for accurate energy forecasts.

OpenSTEF is an Open Source project developed by and for the energy sector to develop more accurate short term forecasting tools to address this challenge. OpenSTEF supports a wide range of energy forecasting use cases. Grid operators can use load forecasts to perform proactive congestion management, to communicate expected energy exchange with other grid operators, or to predict grid losses. District heating operators can apply OpenSTEF for thermal demand forecasting, helping to optimize heat supply and distribution networks. Energy companies can use OpenSTEF to optimize their flexible assets such as renewables or batteries.

The Value of OpenSTEF

Accurate short-term forecasting translates directly into measurable business and operational value across the energy sector. OpenSTEF delivers that value across three dimensions:

Cost Savings

Accurate load forecasts enable grid operators to defer or avoid costly grid reinforcement by managing congestion proactively through demand response, rather than building new infrastructure. Accurate grid loss forecasting reduces operational costs by optimizing procurement. Energy traders, producers and aggregators can use OpenSTEF to maximize returns of flexible assets such as renewables and batteries.

Risk Reduction

OpenSTEF produces probabilistic forecasts with uncertainty bands, helping operators make more confident decisions. By identifying peak moments up to two days ahead, grid operators can proactively steer flexible customers to prevent overloads. Transport forecasts also help coordinate expected energy usage with transmission system operators, reducing network-wide coordination risk.

Efficiency Improvements

OpenSTEF supports the entire forecasting lifecycle and embeds energy domain knowledge throughout, from data preparation to model training and forecast evaluation. Presets provide high-quality forecasts from the get-go. Proven in production, OpenSTEF forecasts congestion across thousands of grid locations at Alliander and optimizes over 40% of Nordic district heating production at Sigholm.

How OpenSTEF Works

OpenSTEF is a Python library that provides complete, end-to-end machine learning pipelines for short-term energy forecasting. To generate forecasts, users input a timeseries dataset of measured load of any type (consumption, generation or a mix). The OpenSTEF pipelines then perform validation and preprocessing of the input data, before using machine learning to combine the historical measurements with external predictors such as weather data and market prices. This way, OpenSTEF deduces the contribution of these external predictors in combination with existing seasonalities for the given timeseries.

The pipelines handle the full workflow, from data preprocessing and feature engineering, through model training, benchmarking and evaluation, to forecasting and post-processing. OpenSTEF produces probabilistic forecasts with uncertainty bandwidths, not just single-point predictions, giving operators a clearer picture of the range of possible outcomes. Forecast horizons cover the short-term range of hours to days ahead as long as accurate weather forecasts are available.

OpenSTEF is built on open standards and its fully Open Source technology stack is designed to be flexible, customizable and extensible. It supports any scikit-learn compatible machine learning model, including built-in support for linear, tree-based and/or gradient-boosted models. In addition, base-learners can be combined through the openSTEF-meta module, and OpenSTEF also supports foundational timeseries models. The library includes built-in, energy-specific feature engineering. For example, it automatically derives solar generation estimates and atmospheric features to improve forecast accuracy. OpenSTEF fits within any IT environment that can serve input data and ingest resulting forecasts – ready to use in downstream applications.

Collaboration

Join our growing OpenSTEF community:

Key Features of OpenSTEF

Operational Focus: OpenSTEF is designed to enable proactive operational energy decisions such as flexibility dispatch.
High Performance: OpenSTEF provides accurate energy forecasts through a wide-range of forecasting strategies.
Energy domain knowledge: embedded from feature engineering to modeling approaches and evaluation metrics.
Customizable Solutions: OpenSTEF offers customizable solutions that can be tailored to any forecasting use case.
Interoperable: designed for effortless integration into your IT environment.

Roadmap

OpenSTEF is actively evolving. For the full roadmap and planned releases, see the OpenSTEF Roadmap on LF Energy Confluence.

OpenSTEF Case Studies

Get Involved

Ready to explore what OpenSTEF can do for your organization?

Whether you are evaluating OpenSTEF for a specific use case, looking to contribute to the project, or simply want to connect with the community, we would love to hear from you. Join the conversation on the LF Energy Slack, attend one of our four-weekly community meetings, or reach out directly by email at openstef@lfenergy.org.

Learn More

Explore how OpenSTEF is driving innovation in grid management and delivering tangible results for utilities like Alliander. Visit the external information sources below to learn more about our initiatives and discover how OpenSTEF can transform your grid operations.

OpenSTEF Videos

24:40

The Value of Open Source Short-term Energy Forecasting Demonstrated Through Real-world Use Cases - Bart Pleiter, Alliander & Jan Maarten van Doorn, Sigholm

Interested to learn more about short-term energy forecasting using open-source technology? This session will showcase the value of openSTEF, LF Energy’s project for short-term energy forecasts, across a range of real-world use cases at different organizations. OpenSTEF provides a valuable portfolio of automated, Open Source machine learning pipelines that provide probabilistic energy forecasts for upcoming hours to days. The pipelines are built to combine measured historical (load) data with external predictors such as weather forecasts and energy prices. We will emphasize the advantages of using Open Source technology and open standards in energy forecasting, along with opportunities for collaboration within the OpenSTEF community. Participants will gain a comprehensive understanding of how OpenSTEF is shaping the future of energy forecasting and ensuring a sustainable and efficient energy grid.

YouTube Video IvUkM9BoZ6A