Simplify – Advanced Machine Learning to support dispatching


Johan Maricq

Artificial Intelligence

Keeping the balance in the grid is essential: the injection of power must always be equal to the consumption to keep the frequency at 50Hz. Due to renewable energy integration, the system is getting more complex and it is becoming essential to forecast imbalances in the short term to keep the grid stable.

The objective of Simplify is to demonstrate that ML models help to formalise decision-making rules, facilitate post-decision evaluation by providing better descriptions of actual situations and displaying correlations between different parameters to raise awareness about the causes of imbalances.


Existing, historical data from nearly two years served as a basis to train Tangent Information Modeler (TIM), an out-of-the-box solution from our partner Tangent Works, and AutoML, an internally developed R-based prototype. The data includes, among others, generated wind / solar power and system imbalances of previous hours. The results were then compared with the forecasts of the existing Excel tool.


The need for new solutions

Due to the increase and fluctuation of renewable energies in the grid and the objective to keep the balancing costs as low as possible, there is a need for better predictions.


Supporting decision-making with AI

The Elia Group ambition is to demonstrate the usefulness of artificial intelligence by supporting control centre engineers in their decision-making.



The results of the project were benchmarked against the currently used Excel “Proto” tool. Therefore, the two following key performance indicators (KPI) were defined:

  • Root Mean Square Error (RMSE), which shows to what extent the actual result deviates from the forecast.
  • Bad Predictions Streaks (BPS), which give a number of sequentially occurring false predictions and the frequency of their occurrence (in %).

By analysing the results, both AI solutions seem to be promising prediction tools for dispatchers, as their KPIs perform a lot better than the existing Proto tool. Nevertheless, TIM’s decisive advantage was its good quality of predictions out of the box and that there is no need to involve internal data scientists. AutoML’s decision-making process is a black box compared to TIM, which makes it hard to identify which of the features play a decisive role in decision-making.

At the moment the project is halted because the CREG (regulator in Belgium) decided that different algorithms for predictions need to be tested first. After a decision has been made, the project will continue.






Visit website

Interested in becoming our next partner?

Partner with us