Forecasts. A topic that is sure to spark discussion, whether you’re an operations manager, an engineer or a business developer. At Aurora by Sigholm (AbS), it’s a key pillar on which our route optimisation is built. Without accurate forecasts, you can’t produce an accurate route plan.
Today, AbS runs forecasts for around 40 per cent of Sweden’s district heating production, with customers in Sweden, Norway and Finland. On the platform, we mainly produce three types of forecasts: electricity prices, weather and load. The load forecasts cover district heating, district cooling, steam and industrial waste heat. All of these serve as input for the schedule optimisation.
Forecasts in production
With around 20,000 forecasts per week, the design and automation of the process are crucial. AbS is a cloud-native application that leverages the scalability of the cloud. Forecast jobs are scheduled hourly and run in parallel for all facilities simultaneously.
The workflow begins with a weather forecast — often based on an external service such as the Met Office, with the option to adjust per site. Load forecasts are then built in parallel. Thanks to our cloud-based infrastructure, the entire forecasting workflow takes just around four seconds, even with thousands of data points. Once all forecasts are complete, the dispatch plan is updated automatically.
Adaptable forecasting engine
Different forecasts have different requirements. Our forecasting engine, which is based on machine learning, can be fine-tuned for each individual plant and load. Multiple forecasting methods can be run simultaneously, and each plant is provided with a model tailored to its operating conditions.
Automatic retraining takes place every week, ensuring that improved models are only activated once their performance has been verified. We can test and compare different models and automatically select the one with the highest accuracy — without any manual effort.
Results
AbS has achieved perfect scores (10 out of 10) in several procurement evaluations and is recognised as the leading platform for CHP optimisation in the Nordic region. Our forecasts achieve an accuracy comparable to the SMHI’s temperature forecasts. Customers report significant operational savings through improved plant utilisation, optimised fuel selection and the elimination of manual forecasting work.
With a robust system that automatically trains models and scales up as needed, we can confidently hand over the operation of forecasting to Aurora by Sigholm — so that we engineers can focus on more enjoyable things: experimenting with new techniques to increase accuracy even further.
The forecasting engine is based on open source code via [OpenSTEF] (https://lfenergy.org/projects/openstef), a framework within LF Energy (Linux Foundation).
