Power System Transients: Impacts of Non-Ideal Sensors on Measurement-Based Applications

The power system is comprised of thousands of lines, generation sources, transformers, and other equipment responsible for servicing millions of customers. Such a complex apparatus requires constant monitoring and protection schemes capable of keeping the system operational, reliable, and resilient. To achieve these goals, measurement is a critical role in the continued functionality of the power system. However, measurement devices are never completely reliable, and are susceptible to inherent irregularities; imparting potentially misleading distortions on measurements containing high-frequency components. This dissertation analyzes some of these effects, as well as the way they may impact certain applications in the grid that utilize these kinds of measurements. This dissertation first presents background on existing measurement technologies currently in use in the power grid, with extra emphasis placed on point-on-wave (PoW) sensors, those designed to capture oscillographic records of voltage and current signals. Next, a waveform “playback” system, developed at Oak Ridge National Laboratory’s Distributed Energy Communications \& Control (DECC) laboratory was used for comparisons between various line-post-monitor PoW sensors when subjected to different high-frequency current disturbances. Each of the three sensors exhibited unique quirks in these spectral regions, both in terms of harmonic magnitude and phase angle. A goodness-of-fit metric for comparing an ideal reference sensor with the test sensors was adopted from the literature and showed the extremes to which two test sensors vastly under performed when compared to the third. The subsequent chapter analyzes these behaviors under a statistical lens, using kernel density estimation to fit probability density functions (PDFs) to error distributions at specific harmonic frequencies resulting from sensor frequency response distortions. The remaining two chapters of the dissertation are concerned with resultant effects on applications that require high-frequency transient data. First, a detection algorithm is presented, and its performance when subjected to statistical errors inherent in these sensors is quantified. The dissertation culminates with a study on an artificial intelligence (AI) technique for estimating the location of capacitor switching transients, as well as learning prediction intervals that indicate the level of uncertainty present in the data caused by sensor frequency response irregularities.