Dynamic dimensioning of frequency restoration reserve capacity based on quantile regression

Frequency restoration reserve capacity is traditionally dimensioned with the help of deterministic criteria or by using probabilistic approaches that determine the capacity for a long period (several months). These static approaches work out quite well with traditional power systems. But increasing shares of intermittent generation introduce higher volatility to today's and future power systems which leads to a more volatile need for balancing. In this paper the main influences on the occurrence of imbalances are identified. Subsequently a new method for the dimensioning of reserve capacities is presented. This method uses quantile regression based on artificial neural networks to forecast the reserve capacities to meet the striven security level. Subsequently the method is tested for the day-ahead dimensioning of frequency restoration reserve capacities in Germany.