Frequency Regulation Capacity Demand Prediction Method Considering Extreme Fluctuations During Scheduling Periods

It is increasingly important to reserve frequency regulation capacity (FRC)for Automatic Generation Control (AGC) units to mitigate the random fluctuations of new energy and increasing loads. The existing methods for quantifying FRC have problems such as unclear FRC correlation characteristics, complex FRC characteristics leading to a large sample demand, and insufficient FRC reserve risks that are difficult to overcome. This paper proposes an FRC demand quantification method to overcome the above challenges based on extreme fluctuation during scheduling periods. Firstly, a Stacked Extreme Learning Machine (SELM) is introduced for learning to avoid the time-consuming training and hyperparameter adjustment process of deep neural networks. Secondly, a fine-grained net load random fluctuation feature screening method based on feature matching degree is proposed. Low dimensional feature vectors that effectively reflect the characteristics of FRC demand are constructed at the data-driven input and output ends, respectively, to reduce the complex mapping relationship between data-driven prediction input and output and improve the accuracy of reliable quantification of FRC demand. Finally, simulation analysis was conducted based on actual data from a provincial power grid in China, and the results showed that the proposed method could provide clear FRC demand and effectively improve prediction accuracy.