Feature Extraction for Post-Disturbance Frequency Prediction Based on Hierarchical Agglomerative Clustering

The prediction of post-disturbance frequency dynamic response plays a significant role in decision making for stability control. This paper proposes a feature extraction scheme for post-disturbance frequency prediction based on hierarchical agglomerative clustering (HAC). The post-disturbance dynamic response process of power system frequency is analyzed, on which the basic input features for prediction are determined. Then, the correlation degrees of each two basic features are calculated to establish a similarity matrix. Moreover, the HAC method is used to extract the representative features. The HAC is conducted with exhaustive centroid-based strategy, and the Elbow's method is used to determine the number of clusters. A case study is presented to validate the feasibility and effectiveness of the proposed scheme.