Ensemble approach for automated extraction of critical events from mixed historical PMU data sets

The scope of this survey is the automated extraction of critical events from mixed PMU data sets without prior knowledge about existing failure patterns. An unsupervised procedure is introduced that identifies arbitrary disturbance types (e.g. voltage sags, frequency drops, oscillations) using an ensemble outlier detection approach. For that, different techniques for signal analysis are used to generate features in time and frequency domain. That approach enables the exploration of critical grid dynamics in power systems. Furthermore new failure patterns can be extracted for the creation of training datasets used for online detection algorithms.