Overcoming Existing Limitations in Electricity-based Artificial Intelligence Applications

Machine-learning algorithms have recently been applied to electrical power problems due to their potential to reduce waste and improve electrical grid reliability, but deployment of existing research is hampered by unrealistic assumptions. My thesis focuses on overcoming these assumptions. Autonomous agents have enormous potential to reduce electrical waste and improve reliability of the power grid. Imagine an ‘energy feedback’ agent at your home or business that identifies electrical waste in real-time, makes recommendations on when to operate different appliances, schedules your dishwasher to run during low-demand hours, and even recommends new appliances you can buy to replace your old, inefficient ones. Such an agent could eliminate billions of dollars of electrical waste every year. Similarly, picture an agent that monitors the power grid, identifies relay misoperations, and corrects them in a fraction of the time required by a human. This agent will have the ability to stop a localized misoperation from creating cascading events, preventing blackouts before they happen.