Development of Machine Learning Tools for Increased Reliability of the Electrical Grid

The electrical grid serves as one of the major lifelines to our society. We have come to expect electricity to be available when we want it. But the system that delivers power to us is very complex and made up of many individual pieces. Maintaining and operating this system reliably is a challenge. Increases in the amount of digital technology on the grid has been both a benefit, in increased amounts of data in order to monitor the system, and a detriment as it provides an increased number of places to attack the grid. This thesis will detail the development of two reliability methodologies. The first seeks to ensure the security of the grid by ascertaining the validity of the data that is being reported by equipment on the grid. The contribution of this section is the formulation of engineered features for a Support Vector Machine that detects spoofed data on the grid. The second proposed methodology resulted in a grid reliability software tool that leverages machine learning clustering algorithms, namely K-means combined with the electrical distance attributes of the system, to determine Voltage Control Areas (VCAs) in a system. These voltage control areas then can help operators approach the data coming from their monitoring systems in a structured way that better aligns with their service zones to operate them more reliably.

Major Advisor: Eduardo Cotilla-Sanchez
Committee: Ted Brekken
Committee: Rakesh Bobba
GCR: Lewis Semprini