Strategies for Addressing Forced Oscillations with the Multichannel Least Squares Mode Meter

This paper analyzes multiple strategies for addressing the presence of forced oscillations (FO) in otherwise ambient data analyzed by the family of multichannel least squares vector autoregressive moving average (LS-VARMA) mode meters. When no action is taken, it is shown that a FO can severely bias mode estimates. To overcome this, an increase in the model order of the VARMA model is used to improve spectral resolution and reduce bias. Alternatively, the LS-VARMA+S mode meter is developed that directly includes forced oscillation information into the model and is able to estimate the oscillation shape of a FO. It is shown that the LS-VARMA+S mode meter is equivalent to the LS-VARMAX mode meter under certain conditions. Each of the strategies are considered through an extensive Monte Carlo simulation study using the miniWECC model. It was seen that all of the strategies provided an improvement, however the LS-ARMA+S approach had the best overall performance while being the most complex.