More accurate forecasts could cut costs for power producers

Production is therefore largely driven by the weather and can be difficult to predict from hour to hour. At the same time, power producers must submit their expected production to the power market one day in advance. Deviations between the forecast and actual production can lead to imbalance costs and expose producers to major price fluctuations in the balancing market.

– Better forecasts can help reduce costs for producers while also improving the balance of the power system, says Kamilla Wergeland of Småkraft, one of the study’s authors and a PhD research fellow at the University of Bergen.

Could reduce costs and risk

For Småkraft’s portfolio of run-of-river hydropower plants, the improvement corresponds to an estimated annual reduction in imbalance charges of around EUR 143,000, or nearly 21 per cent, compared with the current forecasting model. The model also reduced the risk of major losses during periods of extreme prices in the balancing market. This could make operations more predictable for producers and make it easier to keep the power system balanced.

– The results show that combining weather information with recent water-flow measurements can create significant value. We are now working to deploy the model at Småkraft to realise the potential for more accurate forecasts and lower imbalance costs, says Wergeland.

The researchers emphasise that the study demonstrates potential, not a finished solution. In the study, the model used analyses of historical weather as if the weather forecasts were error-free. In operational use, uncertainty in weather forecasts will reduce accuracy. Preliminary tests using actual weather forecasts nevertheless indicate that the model can still outperform the current

Learns from weather and water flow

The model was trained on measurements from the Norwegian Water Resources and Energy Directorate’s monitoring stations and on estimated water flow from 110 run-of-river hydropower plants operated by Småkraft. It was then tested at 30 power plants from November 2022 to October 2025.

Forecast quality improved when the model used the latest water-flow measurements as the starting point for its next calculation. This allows it to make a good estimate of how much water is actually present in the system, including how much snow is lying in the catchment area and how wet the ground is, before estimating production for the following day.

In the tests, the best model had a median forecast error of 23.6 per cent. By comparison, the error was 29.8 per cent for the model Småkraft currently uses and 45.3 per cent for a simple forecast based on production continuing at its previous level. The new model performed better in every season, with the greatest improvement in spring.

The study was conducted by Kamilla Wergeland of Småkraft and the University of Bergen, Ole Wulff of NORCE, and Asgeir Sorteberg of the University of Bergen. The work was supported by Småkraft and the Research Council of Norway through Climate Futures and was published in the journal Renewable Energy.