Singular value decomposition based information retrieval from phasor measurement unit data.

摘要

Electric power systems can display a range of undesirable dynamic phenomena by which acceptable, stable operation may be lost; among these is the "voltage instability" phenomenon. The vulnerability of an electric power system to such phenomenon is typically monitored at intervals of approximately five minutes by state estimators, which are based on a network model to compute the system operating point. The dependence on accurate network parameters and topology, and infrequent monitoring may be regarded as shortcomings in present utility practice. Motivated by this fact, this thesis will propose a conditioning monitor based solely on phasor measurement unit (PMU) data, without the need for network parameters or topology to construct the power flow model and hence suitable for near-real-time implementation.;In a very large electric power system, with very large numbers of measurements over a wide geographic area, computational cost may undermine the goal of near-real time computation. Decomposing a large power system into small partitions is considered, to allow the proposed method to provide dependable results in near-real time. This thesis develops a technique for power system decomposition that blends balanced methods with system coherency. Coherency based power system partitioning with consideration of balanced number of buses in parts may result in improved voltage stability assessment and topology change detection of the proposed method with reduced computation time.;Actual PMU data may be corrupted or missing, which may cause false alarms to operators. PMU data presents low dimensional behavior, which implies that the PMU data array can be estimated as a sum of small number of rank-1 matrices. Exploiting the low dimensional behavior in PMU data, the arriving PMU data can be checked for its consistency with prior observations. The bad/missing data estimation and correction relies on the premise that aggregate power system loads can be decomposed into a slowly varying, deterministic process and faster time scale stochastic process.;This thesis seeks to provide another approach for increasing situational awareness of electric power system by use of a very high resolution data from PMUs, thus exploiting their growing deployment in North America and around the world.