Coherent Clustering Method Based on Weighted Clustering of Multi-Indicator Panel Data

Abstract

The identification of coherent clusters plays an important role in dynamic equivalence and active split control of power systems. The existing coherent clustering methods often adopt a single indicator, e.g. only based on the power angle curve to identify coherent clusters. In addition, in the coherency identification process, the feature extraction is not sufficient, which may cause the problem of inaccurate grouping. In this paper, a coherent clustering method based on weighted clustering of multi-indicator panel data (WCMPD) is proposed. First, the measurements including power angle increment, terminal voltage, and rotor kinetic energy increment from phasor measurement units (PMU) are selected from panel data to reflect the coherence of the generators. Second, the indicator weights and time weights are calculated based on the cross-sectional and time dimension of the panel data. In order to suppress the shortcomings of the coherent clustering method based on Euclidean distance, three distance functions (“horizontal absolute value,” “rate of change at adjacent time points,” and “fluctuation variation degree”) are defined, and then aggregated. At last, the distance matrices among generators are calculated and the coherent generators can be obtained based on the system cluster method. The simulation results on the EPRI-36 bus system and the North China power grid demonstrate that the proposed method has better clustering results than traditional methods.