Comparison of ensemble decision tree methods for on-line identification of power system dynamic signature considering availability of PMU measurements

This paper compares the most commonly used ensemble decision tree methods for on-line identification of power system dynamic signature considering the availability of Phasor Measurement Units (PMU) measurements. Since previous work has shown that the surrogate split method included in classification and regression tree is not good enough to handle the unavailability of measurement signals, more effective methods are needed to be explored. Bagging, boosting and random forest methods are investigated and compared in this work. When evaluating their performance, all possible scenarios of missing PMU measurements are tested for the test network. For each ensemble decision tree model, the result is presented as a probabilistic classification error depending on the availability of PMU signals. The test network used is the 16-machine, 68-bus reduced order equivalent model of the New England Test System and the New York Power System.