Analytics and their applications to power quality and power system data

Multiple monitoring devices deployed throughout transmission and distribution networks enable insightful analysis of power systems' behavior. These devices record signals either at the waveform or phasor levels, and their time resolution range from milliseconds to minutes. This work focuses primarily on the analysis and applications of power quality data at the waveform level. Also, potential applications based on other sources of power system data (namely, smart meters and phasor measurement units) are proposed as a minor contribution of this work.

Modern power quality monitors have the capability of continuously storing the measured waveforms, unlike older devices where only a few data samples were recorded once a disturbance was detected. This evolution created the possibility of new and advanced applications of power quality data analytics. Whereas triggered power quality data have been used only for troubleshooting in the past, triggerless measurements contain more valuable information about the system, which allow event root causes identification and monitoring of equipment health and performance. However, analysis of triggerless power quality data requires parsing through large datasets for finding information hidden in the raw data. In this context, this work proposes two approaches for identifying disturbances and anomalous behavior on triggerless power quality data, which are based either on voltage/current waveforms or rms profiles. The waveform-based detector is a general framework for identifying shape changes between successive cycles of data. On the other hand, rms-based detectors are more suitable for applications related to switching operations, which create a step change in rms voltage/current profiles (such as capacitor switching and voltage regulator operation).

In addition, two power quality data analytics applications are presented. First, a novel method is proposed for accurately characterizing voltage variation events (sags and swells), focused on determining their exact point-on-wave inception and recovery instants. Subsequently, triggered power quality data are employed for condition monitoring of circuit switchers for shunt capacitor banks at a transmission network; this analysis evaluates the performance of pre-insertion resistor/inductor and their switchgear during energizing operations.

Finally, data analytics applications related to alternative power system data sources are discussed. Initially, machine learning techniques are used for predicting in advance voltage magnitudes throughout a distribution network, so that voltage regulation control is enhanced. Next, field data from phasor measurement units are used for illustrating some of the challenges encountered in system frequency estimation; then, a robust frequency estimation technique based on numerical derivatives is proposed for overcoming those limitations.