Computationally efficient weighted updating of statistical parameter estimates for time varying signals with application to power system identification

As power demand continues to grow, the power distribution system is placed under greater stress and moves towards unstable conditions. Recent large scale blackouts such as the ones in western United States in 1996 and the eastern United States in 2003 are visible indications of power grid instability. As a result of these large scale outages, researchers are exploring methods to more closely monitor power grid stability and provide power dispatchers this information in a timely manner.;This dissertation presents three main areas of power grid stability research. The first section looks into weighted updates of statistical parameters for non-stationary signals and their application to the power grid data. The dissertation then moves into an evaluation of 48 hours of western power grid data and the various information it contains. Finally, some initial results from large scale tests conducted on the western power grid are examined.;The first portion of the presented research centers around weighted updates of statistical parameters. In stochastic processes such as the power system data, individual point estimates will often vary over time around an "average" statistical value. The update method involves the use of a weighting function to provide a weighted time-average of several statistical point estimates to produce an estimated value of the current "average" statistical quantity. Using proven mathematical and signal processing methods, a certain class of weighting functions is implemented into a recursive format. The recursive format is applied to estimating power system quantities of interest in a "real-time" implementation.;The 48 hours of western power grid data provides a large, multi-day data set to evaluate trends in the power system data as well as estimate power system parameters at different times of the day. Simple methods of estimation such as mean and variance analysis provide basic information about the power grid data during different periods of the day. These simple evaluation techniques are expanded into modified spectrogram techniques to examine specific frequency trends in the power grid data over the 48 hours. Quantitative parameters of these frequency trends are then obtained using signal processing techniques such as least-squares methods and the weighted updates of parameters discussed in the first section. These various evaluation methods provide insight into the trends and behavior of the power system over a multi-day period.;The final portion of the dissertation examines initial results from large scale tests conducted on the power grid. To help evaluate power grid stability, the Western Electricity Coordinating Council (WECC) conducted a series of tests that involved subjecting the power grid to an impulse-like event and injecting a known "input" to the system. The controlled, deterministic inputs to the power system provide valuable information for fitting models to the power system. A major goal of the different large scale tests is to determine optimal parameters for these injected signals as well as the techniques used to model the power system. With the proper parameters of both the input signal and modeling techniques, continuous, real-time estimates of the power system may be possible and help determine points of instability before a blackout or power system failure occurs.