Precise grid loss for the upcoming day
Kind of data
Database of the transmission system operator
Data mining, cluster analysis, random forest, generalized additive model (GAM) and gradient boosting
Precise forecasting models – 50Hertz
The transmission system operator 50Hertz ensures that electricity in northern and eastern Germany is provided where it is needed – at the customer’s site. But while the electricity is being transported, physical losses occur. These grid losses must be compensated by purchasing the predicted loss quantity via an electricity exchange.
Precise forecasts are indispensable so that the customer knows exactly how much electricity has to be compensated due to grid losses – so 50Hertz uses various statistical models to calculate the possible amount of losses at the previous day. In order to optimize the predictions of grid losses, data sets with the actual grid losses are evaluated in combination with the predicted feed-in of wind and photovoltaic energy.
To identify potential patterns in the data of the observed network losses, eoda uses data mining techniques: Information is filtered from data sets, new variables are extracted and compressed via cluster analysis.
For the customer’s specific application, eoda selected the appropriate method set and tested the methods random forest, generalized additive model (GAM) and gradient boosting for their predictive performance. The models with the highest accuracy were recommended to the customer.
50Hertz can use the models supplied to forecast more precisely how high the grid losses will be for the coming day and thus compensate for the quantity of electricity on the exchange more cost-effectively.
The key is in the data: A seemingly thin database is analyzed with the help of eoda’s data mining – the result is reliable forecasting models that enable the customer to purchase the amount of electricity lost due to grid losses in a predictable and thus cost-effective manner.