A Real-Time Partitioning Method Considering the Spatial and Temporal Distribution Characteristics of Renewable Power System Frequency Response

The integration of renewable energy into the grid significantly influences the spatial and temporal distribution characteristics of system frequency response. Existing fixed partitioning methods struggle to maintain the similarity of frequency responses across different time periods， while real-time partitioning methods often encounter challenges such as partitioning errors and excessive computational demands. To address these issues， this paper introduces a trend and value approximation method to measure the distance between frequency curves. This approach is compared against dynamic time warping， Euclidean distance， and symbolic aggregate approximation distance. Hierarchical clustering and k-means clustering algorithms are then employed to achieve real-time system partitioning. The number of partitions is evaluated using three indices： the silhouette-coefficient （S-C） index， Calinski-Harabas （C-H） index， and Davies-Bouldin （D-B） index. These four distance metrics， two clustering methods， and three evaluation indices are combined to form 24 real-time partitioning models. The optimal model is identified through comparative analysis， which involves calculating trend and value approximation distances， clustering via the hierarchical clustering algorithm， and determining the number of partitions based on the peak S-C index. Finally， simulations are performed in the IEEE 39-bus system， IEEE 118-bus system， and actual power system. The results demonstrate that the proposed real-time partitioning method dynamically adjusts partitioning results in response to different disturbances， locations， and time periods. It ensures similarity in regional frequency responses， enhances the accuracy and speed of system partitioning， and performs effectively in evaluating regional inertia.