OpenSTEF 4 launched at LF Energy Summit Europe 2026 on September 15 in Berlin with a more modular and flexible design for building energy forecasting workflows across different data, software, and operating environments.
In “OpenSTEF 4 Launch: A New Way of Building Energy Forecasting Solutions”, Daan van Es, Chair of the OpenSTEF Technical Steering Committee at Alliander; Jan Maarten van Doorn, Data Scientist at Sigholm Tech AB; and Paula Passet, Data & AI Expert at Sia, presented the new release, Sigholm’s experience migrating from OpenSTEF 3 to OpenSTEF 4, and a generative AI pipeline developed to support documentation for this and future releases. Watch the full session here >
Why OpenSTEF 4 was redesigned
OpenSTEF is an open source Python library for generating energy forecasts from inputs including historical measurements, weather forecasts, and market or price data. It can produce probabilistic forecasts that represent a range of possible outcomes rather than a single predicted value.
Van Es said OpenSTEF 3 had already been used in production across multiple organizations. As the community grew, the project also needed to accommodate a wider range of forecasting use cases, machine learning workflows, and IT environments.
OpenSTEF 4 addresses that need with a more modular design. Different concerns are separated so users can change individual parts of a forecasting workflow and tailor them to their use case.
The release also places a stronger focus on machine learning models and forecasting itself, with less emphasis on surrounding DevOps, MLOps, database, and frontend concerns. Van Es described OpenSTEF 4 as suitable for experimentation, smaller deployments, or integration into an existing software solution as a forecasting module.
More ways to build and test forecasting workflows
The modular design also expands how forecasting workflows can be assembled and evaluated.
Van Es compared the approach to building with blocks: users can combine and adapt different parts of the forecasting workflow according to the problem they are trying to solve.
Capabilities presented during the session included:
- Configurable forecasting workflows that make it easier to change preprocessing or model components for a specific use case.
- Model ensembles that allow multiple models with different objectives to be combined, such as pairing a general forecasting model with one focused specifically on peak conditions.
- Foundation time series model support that provides another forecasting approach through the broader OpenSTEF interface.
- Dedicated backtesting that can simulate historical production workflows, including repeated training and forecasting, so teams can compare modeling setups, preprocessing choices, and input data before deployment.
Van Es emphasized that OpenSTEF does not assume one forecasting approach is universally best. The appropriate setup depends on factors including the use case, available data, and forecasting requirements.
Sigholm migrates its district heating forecasts to OpenSTEF 4
Jan Maarten van Doorn presented Sigholm’s use of OpenSTEF for district heating demand forecasting in Sweden. Those forecasts provide input to software used to optimize heat production.
In Sigholm’s use case, the forecasting task differs from the congestion-focused electricity examples discussed earlier in the session. District heating demand is strongly influenced by weather, and Sigholm forecasts the broader demand curve over a longer horizon to support production and fuel planning.
Sigholm had already been using OpenSTEF 3 and migrated its forecasting workflow to version 4. Van Doorn said the migration went well and highlighted the new backtesting capabilities and modular architecture.
Because OpenSTEF 3 models cannot be used directly in version 4, Sigholm retrained its models as part of the migration. The team used OpenSTEF 4’s backtesting capabilities to check forecasting quality before using the updated models operationally.
Van Doorn also said the more modular design allowed Sigholm to reduce coupling between components in its software architecture.
Sigholm also developed an operational inference selection improvement that Van Doorn said the team plans to contribute upstream to OpenSTEF.
Keeping documentation aligned with the code
Alongside the OpenSTEF 4 release, Passet presented a new approach to maintaining project documentation.
She described a generative AI pipeline developed to help documentation keep pace with changes to OpenSTEF. The pipeline is deployed on AWS and can be triggered by activity in the project’s GitHub repository, including a new release.
The generation process uses a knowledge base populated with project code, code examples, and docstrings. Passet also emphasized a human-in-the-loop design: suggested documentation changes must be checked, approved, and, when necessary, edited by OpenSTEF maintainers before they are merged.
For the OpenSTEF 4 release, Passet said the pipeline generated 12 pages for the user guide. It also triggered automatically on five smaller releases following version 4.0.
The documentation work also included diagrams and a tutorial video showing new users how to run their first OpenSTEF forecast.
Passet said the team intends to extend the approach to other projects, with Power Grid Model in the onboarding process at the time of the session.
OpenSTEF 4 is designed for different forecasting use cases
Taken together, the session presented OpenSTEF 4 as a set of forecasting building blocks that organizations can adapt to different models, deployment patterns, and software environments.
Van Es emphasized flexibility in how forecasting workflows are assembled and tested. Van Doorn showed how those changes applied to an existing district heating implementation migrating from version 3. Passet showed how the project is also addressing the documentation needed to support continued development and adoption.
The result is a release designed to let users shape forecasting workflows around their own data, systems, and use cases rather than requiring a single prescribed approach.
Watch the session
Watch the full session, “OpenSTEF 4 Launch: A New Way of Building Energy Forecasting Solutions.”
Frequently asked questions
What is new in OpenSTEF 4?
OpenSTEF 4 introduces a more modular design focused on machine learning and forecasting. The release supports configurable workflows, model ensembles, foundation time series models, and dedicated backtesting so users can tailor forecasting approaches to their data and use cases.
How are organizations using OpenSTEF?
The session highlighted OpenSTEF use across multiple energy settings. Alliander uses it for electricity-grid forecasting, Sigholm uses it for district-heating demand forecasting, and E.ON uses it in a redispatch process.
How is documentation maintained for OpenSTEF 4?
Sia presented a generative AI documentation pipeline that can be triggered by GitHub repository activity, grounds generated documentation in project code, code examples, and docstrings, and requires maintainer review and approval before changes are merged.
About OpenSTEF
OpenSTEF (Open Short-Term Energy Forecasting) is an open source machine learning pipeline for short-term energy load forecasting, developed by Alliander and hosted at LF Energy. It produces probabilistic 48-hour-ahead forecasts at individual substation and feeder locations for operational congestion management. Alliander uses OpenSTEF in production at hundreds of grid locations; Sigholm uses it as the forecasting engine for its Aurora platform, which optimizes approximately 40% of Sweden’s district heating production. RTE also contributes to the project.
About LF Energy
LF Energy is the Linux Foundation’s neutral home to build the digital foundation for energy together. We host more than three dozen open source projects used to plan, operate, and modernize energy systems. The portfolio spans from transmission system modeling to substation virtualization, smart meter integration to EV charging. We provide the trust layer: neutral governance, legal frameworks, security practices, and community of practitioners that turn open source software, standards, and data into infrastructure the industry can depend on for decades. Members include utilities, energy companies, vendors, researchers, and technology providers, aligned to deliver affordable, reliable, safe, and clean energy. For more information, please visit lfenergy.org.
About The Linux Foundation
The Linux Foundation is the world’s leading home for collaboration on open source software, hardware, standards, and data. Linux Foundation projects are critical to the world’s infrastructure including Linux, Kubernetes, Node.js, ONAP, PyTorch, RISC-V, SPDX, OpenChain, and more. The Linux Foundation focuses on leveraging best practices and addressing the needs of contributors, users, and solution providers to create sustainable models for open collaboration. For more information, please visit linuxfoundation.org.
AI Disclosure
This post used artificial intelligence tools for research, structural assistance, or grammatical refinement. The final content was reviewed, edited, and validated by human contributors to LF Energy to ensure accuracy and alignment with our community standards. We remain committed to transparency in the use of generative technologies within the open source ecosystem.
Last updated: October 15, 2026