At LF Energy Summit 2024 in Brussels, Daan van Es from Alliander delivered a session on “Net Congestion Forecasting.” As a product owner in short-term forecasting, van Es shared valuable insights into the critical challenge of managing congestion in the energy grid, a problem that has gained prominence across Europe, particularly in the Netherlands. A summary of the session follows, and the full video is posted at the end.
The Role of Grid Operators
Van Es highlighted that Alliander, as a Distribution System Operator (DSO), plays a pivotal role in ensuring efficient energy distribution without engaging in production or storage. The growing strain on the Dutch grid, largely due to increasing renewable energy sources like solar and wind, has led to frequent media coverage about net congestion issues. For example, van Es pointed out that the grid is becoming overburdened, with lengthy waiting queues for connections and energy installations, such as solar and wind parks, being temporarily turned off at peak moments.
Long-Term vs. Short-Term Solutions
The presentation outlined two broad approaches to addressing grid congestion:
- Physical Expansion: This involves upgrading grid capacity by installing more cables and transformers. However, van Es noted that this is a slow, resource-intensive process, involving significant delays due to permits and material shortages.
- Smart Solutions: Alliander is focusing on managing energy flows more flexibly. By understanding the peaks in energy consumption or generation, particularly from renewable sources, operators can optimize existing infrastructure. For instance, during periods of low grid usage, there is still room to balance load by shifting consumption or generation.
Forecasting for Net Congestion
Alliander’s system, called “Grid as a Service,” includes several components for managing grid congestion. At the heart of this system is a short-term forecasting module, which predicts energy demand and supply for specific points in the grid. This forecasting is probabilistic, offering a range of expected values and uncertainties, which helps operators prepare for possible congestion scenarios.
Forecasting is key for identifying overloads and taking action before grid capacity is exceeded. Alliander uses historical data on energy loads and weather conditions to train its models, ensuring they accurately reflect current grid dynamics. Van Es explained that this predictive capacity allows DSOs to manage grid congestion proactively, such as entering into agreements with renewable energy providers to temporarily reduce output at peak times.
Challenges in Forecasting
A key challenge, according to van Es, lies in the unpredictability of renewable energy sources like solar and wind. Both types of energy generation are subject to varying weather conditions, making them more erratic than traditional consumption-based loads. Van Es emphasized that accurately forecasting net congestion requires balancing the unpredictability of generation with flexible grid management strategies.
Machine Learning and Open Source Tools
Alliander leverages machine learning for forecasting, using the open source tool OpenSTEF (short-term energy forecasting). OpenSTEF is a complete software stack designed for data validation, feature engineering, training models, and generating forecasts. Van Es explained how the system can easily be adapted and trained using historical data, making it a powerful tool for grid operators looking to mitigate congestion.
However, van Es cautioned that machine learning models are only as good as the data they are trained on. As the grid evolves – both in terms of consumption patterns and renewable energy installations – forecasting models must be continuously retrained with up-to-date information. This dynamic nature makes forecasting more complex, but also essential for managing the growing penetration of renewables.
Practical Use Cases and Precision
Van Es shared examples of how Alliander’s models have been deployed in real-world scenarios, discussing how the company balances accuracy and usability in its forecasts. By converting forecasting problems into classification tasks, Alliander ensures that grid operators can act swiftly to prevent overloads. The precision of forecasts is also crucial: high recall rates ensure operators can detect peaks in energy demand, helping to maintain grid stability.
Open Access to Forecasting Models
One of the highlights of van Es’s talk was the open source nature of Alliander’s forecasting tools. By making OpenSTEF available to the broader energy community, Alliander is contributing to global efforts to improve grid flexibility and resilience. The software, which is available on GitHub, provides example notebooks and reference implementations for users interested in applying these models to their own grid challenges.
