Data Mining Application to Power Grid PMU Data

摘要

The reliability of the power system is of vital importance to daily life. Power outages, either small or big, cause economic loss and inconvenience. The effort to better understand the behaviors of the power grid has a long history. Several years ago, engineers and researchers started to modernize the national power grid in the United States by moving it into the next generation called SmartGrid. The massive installation of Phasor Measurement Units (PMUs) is the highlight of this movement. Compared to the sensors from the last generation in the power grid, PMUs can produce more accurate and rapid information regarding the state of the grid.;Data from PMUs are usually 30 samples to 60 samples per second and contain both voltage magnitude and angle measurements. The data is high-dimensional with hundreds or thousands of signals and is highly correlated across signals. One advantage of PMU data is that it is time aligned throught GPS. The biggest challenge in the use of PMU data is the massive amount of data, which causes difficulties for storage, pre-processing, analysis, and visualization. A whole year of PMU data can be several terabytes depending on the number of signals in the data. In fact, the information provided by PMUs is now so big that it is difficult for scientists to handle or easily comprehend.;Meanwhile, many exciting accomplishments have been seen in various fields using data mining. Data mining has become increasingly important with the appearance of various kinds of big data. The power grid data analytics is a good example of such a big data problem. There has been an increase of data mining applications in the power systems research field in recent years, partly due to advancements in data mining. There has been much work on this topic in the last 10 years.;This research work, which has been done at the University of Wyoming and Pacific Northwest National Laboratory, consists of data analytics on simulated PMU data from the MinniWECC system and real PMU data from the Western Electricity Coordinating Council (WECC) system for the 2008 to 2009 and 2016 to 2017 operating seasons. This work contains measurement-based power system offline studies involving event detection, event classification, abnormal operating conditions, and potential online applications.;The simulated study discusses the implementation and performance of various machine learning algorithms for classifying power system event types and event locations. A simple feature extraction method is applied. The contribution of this study is to demonstrate how data mining techniques can be used to incorporate information from PMU data to assess the system condition.;Moreover, data mining techniques are applied to historical data consisting of PMU measurements from WECC from June 2008 to June 2009. The main objective is to classify abnormal and normal power grid modal behavior of the WECC interconnect at the daily scale. The data is transformed to the frequency domain to represent operating conditions of each day. A closer investigation of misclassified days is also conducted to look at abnormal system behaviors at the hourly scale. The research contribution of this study is the application of data mining techniques to power grid data in the frequency domain to identify various power system events, especially large scale events both in size and in duration.;The third part of this dissertation extends the findings in the simulated study and applies updated methodologies to PMU data from October 2016 to May 2017. This work involves training machine learning algorithms to detect and classify power system events in the time domain. Different machine learning algorithms are applied and a new algorithm is developed to enhance the final algorithm. The results show that the proposed algorithm can successfully detect and classify power system events at high accuracy in under one second. This research demonstrates the potential for an on-line application of achieving near real-time power system situational awareness.