Quantification for Frequency Regulation Speed Requirement Based on Interval Prediction of Maximum Short-Term Power Fluctuation

The construction of a new power system, with renewable energy as its main component, is essential for achieving our country's "dual carbon" goals. However, the inherent randomness and fluctuations associated with renewable energy sources pose significant challenges to the stable operation of the power grid. By establishing boundary conditions for day-ahead power system dispatch decisions, we can better manage the short-term fluctuations in system net load during the dispatch period. To address this issue, this paper proposes a quantitative method for determining the power system's frequency regulation speed requirements based on the predicted maximum short-term fluctuations in net load power. First, a sample eigenvector is constructed to forecast the maximum short-term power fluctuations of the net load. Then, training samples are selected based on weighted Euclidean distance, and input features are filtered according to the weighted training samples. Second, leveraging the deep neural network's (DNN) strong nonlinear feature extraction capabilities, we develop a prediction model for the maximum fluctuation point of short-term power. Finally, using the DNN point prediction model in conjunction with the adaptive bandwidth kernel density estimation (ABKDE) method, a model for predicting the maximum short-term fluctuation interval of net load power is constructed. The system frequency regulation speed requirement is then quantified based on the upper limit of this prediction interval. Simulation results from a numerical example demonstrate that the proposed method effectively captures the actual maximum short-term power fluctuations in net load. Furthermore, the frequency regulation speed requirement, determined by the upper limit of the prediction interval, ensures the system's setting adequacy.