Model order sensitivity in ARMA-based electromechanical mode estimation algorithms under ambient power system conditions

This paper presents the results of a model order sensitivity study performed on ARMA-based electromechanical mode estimation algorithms under ambient power system conditions. Both single- and multichannel versions of the Yule-Walker (YW) and Least Squares (LS) approaches are considered along with a Recursive Maximum Likelihood (RML) estimator. An extensive Monte Carlo study was performed using a reduced-order model of the Western United States power system, where it was seen that while all of the algorithms had similar accuracy in their best case scenarios, the YW algorithms were far less sensitive to changes in model order. These results are supported with a study using data measured from the actual WECC system. Additionally, the study revealed a numerical instability inherent to the RML approach that has identified an important area of future work.