Data-mining-based approach for predicting the power system post-contingency dynamic vulnerability status

摘要

This paper proposes a data-mining-based approach for predicting the power system post-contingencyrnvulnerability status in real time. To this aim, the system dynamic vulnerability regions (DVRs) are firstrndetermined by applying singular value decomposition, by means of empirical orthogonal functions (EOFs),rnto a post-contingency data base obtained from phasor measurement units (PMUs) adequately locatedrnthroughout the system. In this way, the obtained pattern vectors (EOF scores) allow mapping the DVRsrnwithin the coordinate system formed by the set of EOFs, which permits revealing the main patternsrnimmersed in the collected PMU signals. The database along with the DVRs enable the definition, training,rnand identification of a support vector classifier (SVC) which is employed to predict the post-contingencyrnvulnerability status as regards three short-term stability phenomena, that is: transient stability, short-termrnvoltage stability, and short-term frequency stability (TVFS). Enhanced procedures for feature extractionrnand selection as well as heuristic optimization-based parameter identification are proposed to ensure a robustrnperformance of the SVC. Numerical results, obtained by implementation of the proposed approach on tworndifferent size test power systems, demonstrate the methodology viewpoint and effectiveness. Copyright ©rn2014 John Wiley & Sons, Ltd.