Daily electricity demand forecasting with artificial intelligence

The amount of electricity generated in Germany sometimes differs from actual demand – which can be higher or lower than expected. To stabilise the grid and energy supply, so-called balancing capacity has to be made available. The Federal Ministry for Economic Affairs and Energy has allocated funding of €570,000 to a research project with the goal of accurately forecasting nation-wide balancing capacity on a daily basis.

Strommasten bei Sonnenuntergang© BMWi/ Holger Vonderlind

For many years, electricity demand forecasts were based on experience and covered a period of three months. However, electricity generation in Germany has changed tremendously over the past decades. The share of renewables in gross electricity consumption has steadily increased and reached 27.8 per cent last year. To account for this development, researchers at the Fraunhofer Institute for Wind Energy and Energy Systems Technology have developed a new forecasting method as part of their research project "Dynamic forecasting of balancing capacity demand". The forecast takes many fast-changing factors into account: Weather conditions and temperatures at specific times, which influence the amount of energy fed into the grid from PV systems and wind-powered installations.

Forecasts of balancing capacity at quarterly intervals have repeatedly proven unreliable in the past. One striking example: With the unusually cold winter at the beginning of the year 2012, electricity demand was unusually high in Germany. As a result, balancing energy demand was much higher than expected and grid stability was in jeopardy. In future, such unpredictable weather conditions can be directly factored in the new forecasting method. This also applies to situations where electricity generation suddenly soars due to long periods of sunshine or wind.

The research project found that the new and more dynamic forecasting method is more accurate in determining the daily balancing capacity demand than the previously used static approach. Consequently, less balancing capacity is needed, resulting in high cost savings and improving energy security and grid stability.

The researchers’ forecasts are based on artificial intelligence, an artificial neural network, which is fed information and can learn over time. Historical time series about all the factors that could influence the forecasts were fed into the system, allowing it to create the necessary links. The forecasting method also includes a corrective function, allowing daily updates.

Context: In general, balancing capacity is used in situations where electricity generation and consumption are not balanced. Without balancing capacity, the electricity grid would become unstable or even break down. Balancing capacity is available in varying qualities:

  • Primary balancing capacity is available fastest. Electricity-generating installations have to be able to provide primary balancing capacity within 30 seconds.
  • Primary balancing capacity is replaced by secondary balancing capacity after a maximum of five minutes.
  • If electricity supply and demand are imbalanced for longer time periods, a minute reserve comes into play, which can be activated within 15 minutes.

In the past, balancing capacity was mostly provided by large-scale power plants and pumped storage power plants. For a few years now, biogas plants have also been used.

Further information