Discovery of False Data Injection Attacks on Power Grid Frequency Controllers with Reinforcement Learning

Research output: NRELPoster

Abstract

While inverter-based DER (distributed energy resources) are instrumental to integrating renewable energy into the power grid, they reduce the grid's mechanical inertia, thereby increasing the risk of frequency instabilities. To compensate for frequency instability risks, the grid must also undergo a transformation to include digital technologies that allow for two-way communication between the utility and customers. The current and future state of the power grid allows for building a cleaner energy landscape. However, the grid may also become vulnerable to novel cyber threats. To preemptively protect the power grid against elaborate cyber attacks, we propose to discover potential threats via reinforcement learning. In this work, the focus is on studying false data injection attacks that target the control logic of frequency controllers. We show that a reinforcement learning agent can successfully discover how to best inject false data into linear droop controllers.
Original languageAmerican English
PublisherNational Renewable Energy Laboratory (NREL)
DOIs
StatePublished - 2023

NREL Publication Number

  • NREL/PO-2C00-87148

Keywords

  • false data injection
  • frequency controller
  • reinforcement learning

Fingerprint

Dive into the research topics of 'Discovery of False Data Injection Attacks on Power Grid Frequency Controllers with Reinforcement Learning'. Together they form a unique fingerprint.

Cite this