| Rising numbers of Phasor Measurement Units (PMUs) are being installed in the North American power grid. These devices allow for an increased situational awareness of the power grid by recording synchronized measurements of power system variables at a rate of up to 60 sample points per second.;PMU data drops are, however, prevalent and affect the reliability of this technology. High data accuracy and reliability is key to the widespread use of these devices, motivating an industry interest in developing robust PMU data recovery algorithms. Given that the power system is an interconnected network, an approach was previously developed which makes use of low-rank matrix completion algorithms as a tool for solving the PMU missing data recovery problem. This approach can leverage the spatial characteristics of a dense PMU network as well as its temporal characteristics.;This thesis evaluates two existing low-rank matrix completion algorithms, the Singular Value Thresholding (SVT) algorithm and the OnLine Algorithm for PMU data processing (OLAP). The SVT algorithm iteratively solves a modified version of the traditional nuclear norm minimization problem. The OLAP algorithm is a real-time data recovery method which allows for the dimension of the recovered subspace to vary with time, an innovation which is key in the recovery of PMU data during grid disturbances.;The two algorithms were used to recover missing data from real New York State (NYS) PMUs during grid disturbances. Frequency, voltage magnitude, voltage angle, current magnitude, and current angle recovery were performed separately. The results of the recovery were evaluated, and the two approaches compared.;Additionally, a disturbance detection algorithm was developed. This algorithm also leverages the low-rank matrix properties of PMU data to detect disturbances as they propagate through the grid, triggering for both an increase in frequency derivatives, as well as an increase in singular values. |
