2 years ago
More efficient simulation of water flows in hydropower plants

Calejo has helped Uniper to develop an AI model of hydropower production in four hydropower plants. The result was a forecast and a fast and scalable simulation, leading to higher revenues and less resource-intensive production and maintenance.

Sydkraft Hydropower, a subsidiary to Uniper, operates four hydropower plants in a section of the Åsele River between Hälla village and Junsele in the northern part of Sollefteå municipality. Between each power plant, there are reservoirs that are used for regulation for power production in the power plants.

All planning of power production is based on how much water is available in the reservoirs and for this available water volume based on measurements of the reservoir surface is fundamental. A good volume determination facilitates the possibility of following an optimal timetable.

Forecast of water flows Calejo was commissioned by Uniper to use AI to determine the volume of the three water reservoirs at the centimeter level over time. In dialogue with Uniper, an AI model was created, which, based on measured data, was able to forecast water flows and determine uncertainties in these.

The project was able to establish that an AI-based model for production planning and maintenance gives better results than ordinary classical, statistical analysis.

- Our AI model creates an automatic self-learning business understanding, which can handle very complex modeling, automatic control at station and planning level, satisfactory production planning and in the long run also enables automated station operation, says Leonard Johard, CTO in Calejo and responsible for the project.

Successful result The modeling result provided a fast and scalable simulation over relevant horizons with optional time resolution.

- The more important systematic errors were reduced, while more complex automatic magazine models reduced the average error in the surface forecast by more than 30 percent, a figure that was judged to be significantly higher in the future as new data teaches the model to be even more confident in its forecasts. Leonard Johard.

Calejo