To stabilize the increasing random fluctuations of new energy and load

it is increasingly important to reserve the frequency modulation capacity by AGC (automatic generation control) units. Frequency regulation capacity needs to effectively cover the extreme values of random fluctuations in the net load curve during the dispatch period to meet the real-time power balance demand of the power system. The existing method for frequency regulation capacity requirements quantificaiton suffers from the unclear correlation characteristics of frequency regulation

the large sample-size demand due to the complex frequency regulation capacity characteristics

and the risk of insufficient frequency regulation reservation. At present

the industry still determines the frequency regulation capacity requirements based on experience. In order to overcome the above problems

this paper proposes a new method for quantifying frequency regulation capacity requirements based on credible prediction of fluctuation extremes in scheduling periods. Firstly

for small sample scenarios

a Gaussian process model is constructed based on the Gaussian kernel function to predict the extreme value of net load fluctuation in the dispatch period. Secondly

a high-resolution screening method for random net-load fluctuation features based on feature matching degree is proposed. The low-dimensional feature vectors of frequency regulation capacity requirement are constructed to reduce the complex mapping relationships between input and output in data-driven prediction and improve the quantification accuracy of frequency regulation capacity requirements. Thirdly

in order to reliably quantify the frequency regulation capacity requirements

a method based on chance constraint is proposed

which introduces chance constraint to evaluate the uncertainty band of its prediction error. The Monte Carlo simulation method is used to generate scenario-sets mapped to the corresponding uncertainty band

and the frequency regulation capacity requirements are reliably quantified at a certain confidence level. Finally

simulation analysis is carried out based on the actual data of a provincial power grid in China. The results show that the proposed method can accurately quantify the frequency regulation capacity requirements and ensure its effectiveness within a given confidence interval.