In the control area of the German transmission system operator (TSO) 50Hertz in northern and eastern Germany, the installed capacity of renewable energies is particularly high. Since renewable energy in Germany is typically generated at a considerable distance from the centres of consumption and thus must be transmitted over long distances, the transmission losses for 50Hertz are correspondingly high. In 2018, TSCNET shareholder transmitted a total of 117TWh of electricity and the losses amounted to 2.5TWh which corresponds to 2%.
TSOs must compensate for these losses through costly feed-in measures. 50Hertz spent around €70m in 2018 on compensatory electricity and has a keen interest in predicting losses more accurately and purchasing electricity for compensation at lower cost on the electricity market. Therefore, 50Hertz has developed a new prognosis model based on Artificial Intelligence (AI).
The previous model was based on forecasts for the expected feed-in of wind and solar power and a comparison with similar days in the past. For the new model, 50Hertz provided a much larger database: At 70 different locations in the 50Hertz control area, data on the feed-in of renewables as well as on the amount of wind, insolation and temperature are collected every quarter of an hour. With this enormous quantity of data, the artificial neural network is to calculate a transmission loss forecast for the following day, also with an accuracy of 15 minutes. The forecasts are then compared with the actual grid losses and the algorithm is automatically adjusted. This process is repeated thousands of times and the forecasts become more and more accurate.
The neural network has been in the test phase since the end of June 2019. Since then, the data has been cleansed and a database has been built so that the AI algorithm has been continuously improved and highly reliable prognoses can now be created. Since 9 December, the model is fully applied and used operationally at 50Hertz.