Combining least mean squares adaptive filter and auto-regressive block processing techniques for estimating the low-frequency electromechanical modes in power systems

A variety of techniques have been developed for estimating the low-frequency electromechanical modes of power systems based on the analysis of complex system models, ring downs from a system disturbance, and noise injection signals. This work uses a combination of adaptive filtering and block processing techniques to estimate these modes from ambient power system data. Current methods of using these techniques separately have seen constraints arising with the convergence time and variability of the estimates for adaptive algorithms and the large blocks of data required for block processing. This paper investigates possible ways of overcoming these constraints by evaluating the performance of the least mean squares (LMS) adaptive filtering algorithm, taking into consideration the step size parameter (mu) and the initial weight vector estimate from auto regressive (AR) block processing. This technique is applied to simulated data containing a stationary low-frequency mode generated from a 19-machine test system and to actual ambient power system data recorded from the western North American power grid in June 2000. The frequency and damping factor estimates from LMS alone and in combination with AR are compared with the actual modes (test system) and with over-determined AR (full block of ambient data) to measure the improvement in convergence time and the variability of the estimates