Robust Event Cause Analysis in Power Grids using Machine Learning Algorithms
Loading...
Authors
Niazazari, Iman
Issue Date
2019
Type
Dissertation
Language
Keywords
Adversarial Machine Learning , Cyber-physical Power System , Event Analysis , Feature Learning , Machine Learning
Alternative Title
Abstract
Power grids are composed of generation, transmission, distribution and customer level assets along with protection, monitoring, and control equipment that are well-coordinated and operated to deliver sinusoidal voltage and current waveforms with desirable magnitude and frequency. However, faults, abnormal events, such as load or generation outages, grid equipment failures, assets malfunctioning, or lightning strikes occur that prevent the power grids from delivering the desired quality of service to the customers. These events frequently occur in the grid and protection devices usually isolate the faulty and malfunctioning sections of the grid to prevent further damage to the grid asset and equipment, and to prevent the propagation of disturbances to other sections of the grid. Fast and accurate detection and classification of abnormal events will, therefore, lead to a more accurate root cause analysis of failures, a quicker system restoration process after system disturbances, and less adverse impacts from socioeconomic, national security, and public health perspectives. Therefore, establishing an accurate event diagnostics (i.e., detection and classification) framework to extract useful information such as the cause or location of events is of utmost importance. On the other hand, disruptive and abnormal events may not cause immediate and direct failures on the grid. However, they are potential sources for permanent failures over time. Therefore, providing a framework for detecting and distinguishing these events from each other provides electric utilities with a comprehensive post-event analysis mechanism that can be used for preventive maintenance scheduling of the equipment exposed to the adverse events, or preventive actions prioritization to mitigate the potential future adverse impacts. This dissertation will present novel and robust data-driven frameworks for event cause analysis in power grids (distribution and transmission) based on the state-of-the-art measurement devices equipped with global positioning systems (GPS).