Wind Power Forecasting

With no way of efficiently storing electricity on a large scale, grid managers have to match supply and demand at all times to ensure the grid stays safe and reliable. Unfortunately, planning the grid has become substantially harder with the inclusion of intermittent energy sources, such as wind power, into the energy mix. Directly dependent on weather conditions, wind power is extremely difficult to forecast and production planning is notoriously difficult.

Windmills and green meadow in northern Denmark.

Global weather models are run every six hours, trying to model the weather patterns on a global scale. In order to generate a precise power forecast though, we need accurate, local weather forecasts as well - even down to each individual wind park. The challenge however, is even more complex. Local terrain also affects the weather and thus, the location of every single turbine in the park have to be considered. To further increase complexity, turbine efficiency can change over time and there will be occasions when they need maintenance.  

To help producers and grid managers face these challenges, Powel is providing state-of-the-art machine learning algorithms for improved power forecasts. The machine learning models utilise live data from the wind parks, coupled with high quality weather forecast to learn global and local weather phenomena. In addition to providing precise forecast, the model is quick to respond to changes in weather conditions. As it has already been trained on similar historical situations, it can quickly produce an updated forecast incorporating the changes in the behaviour of the wind.

Benchmarked against traditional forecasting systems, our models normally reduce the forecasted errors with 10-20% in the short term, making our services optimal for short-term planning, intraday trading or reduction of imbalance cost. In Norway, 10-20% improvement in imbalance cost could be translated as approximately NOK 1 million / Euros 100,000 per GW capacity per year.

The machine learning model is set up to learn from different data sources, and precision will increase the more data it has available. However, the only requirement is historical and live production data from the relevant wind park. 

To be able to deliver stable services with high uptime, our wind power services are managed with Kubernetes technology in Microsoft Azure.

  

We track our performance live with Power BI

dashboard with graphs and charts

 

Do you want to know more about Wind power forecasting? 
Get in touch with us to discuss your needs and the possibility of a trial today.