Hybrid clustering-based bad data detection of PMU measurements

Phasor measurement units (PMUs) have been widely deployed in power grids, while the bad PMU data problem threatens power system monitoring and control. This paper first gives the objective of the bad PMU data detection and gives an illustrative bad data instance. Then, the time-series PMU data of neighbouring buses are cast as a two-dimensional diagram, of which the spatio-temporal correlation analysis is performed to design the normal and outlier data detection problem. Three clustering methods, including linear regression, density-based spatial clustering of applications with noise (DBSCAN), and Gaussian mixture models (GMM) are ensembled for bad PMU data detection. Moreover, the statistical analysis and bound modification of data clustering are developed to further improve the detection accuracy. Finally, the procedure of the two-stage bad PMU data detection is given, which consists of ensemble learning and modification. The proposed hybrid clustering-based bad data detection is unsupervised and is applied to online bad PMU data detection with a short computation time. Visible and numerical case study results validate the outperformance of the proposed method.