Unsupervised Machine Learning Approach to Enhance Online Voltage Security Assessment Based on Synchrophasor Data

Abstract

The accuracy and reliability of the Q/V sensitivity for voltage security assessment is influenced by the outliers present in the calculation results. An unsupervised machine learning approach, empirical- cumulative- distribution- based outlier detection (ECOD), is introduced in this letter to detect and eliminate outliers to address this issue. A comparison of the results with those of the proposed approaches on the standard test power system CSEE-VS demonstrate that, compared with advanced outlier detection algorithms, ECOD can eliminate outliers from the Q/V sensitivities with higher accuracy and less computation time and realize online voltage security assessment with superior accuracy and reliability.