Electric grid protection through low-cost sensors, machine learning
Connecting state and local government leaders
Researchers at Lawrence Berkeley National Laboratory have developed a relatively inexpensive method for on-the-fly detection of attacks on power grids.
Because virtually all industries and sectors rely upon electric power, the security of the electric grid often dominates cybersecurity discussions about critical infrastructure.
Researchers at the Department of Energy’s Lawrence Berkeley National Laboratory are already doing their part to secure the nation’s vulnerable electricity grids. They have developed a relatively low-cost method for on-the-fly detection of attacks on power grids.
Like some other researchers, the Berkeley team has focused on monitoring phasor measurement units (PMUs), sensors that are installed in fixed locations throughout the grid and send data about the status of the grid at a rate of 30 measurements per second to a control center.
"The idea is if we could leverage the physical behavior of components within the electrical grid, we could have better insight in terms of whether there was a cyberattack that sought to manipulate those components," said Sean Peisert, a computer scientist in Berkeley Lab's Computational Research Division and a cybersecurity expert. “These devices provide a redundant set of measurements that give us a high-fidelity way of tracking what is going on in the power distribution grid," he said. Researchers can look at the measurements to detect anomalies, or they can compare the sensor readers with what the equipment reports to find discrepancies that could indicate attacks against components in the power grid.
The problem, according to Peisert, is that PMUs are expensive and, as a result, are generally deployed only at high-voltage substations. Relying exclusively on measurements from those sensors may result in an attack going undetected for some time. An attacker could manipulate readings from a single sensor and damage the grid, he said.
To cast a wider net, the Berkeley team is turning to micro-PMUs, smaller and less-expensive units that take four times more measurements than the current sensors. The micro-PMUs can be deployed throughout the grid, and by taking 120 measurements per second can provide a higher resolution picture of the status of the grid and greater redundancy.
Data from the micro-PMUs is combined and then sent to existing SCADA (supervisory control and data acquisition) systems widely used by utilities to provide real-time feedback.
The Berkeley team also modified an existing algorithm -- the cumulative sum algorithm, which was created in 1954 for sequential analysis of the data and automated anomaly detection -- to detect abnormal behavior in the power grid.
The machine-learning algorithm learns to distinguish between abnormal and normal behavior "by detecting changes in the physical environment, such as current magnitude and active and reactive power," said Ciaran Roberts, an energy systems engineer at Lawrence Berkeley. "All the computing is done in real time during the physical data collections, and the algorithms are designed to run in real time."
The three-year project, launched in 2015, is nearly ready to move the technology to the field. The units were developed by Power Standards Lab as part of a project led by UC-Berkeley and funded by the Department of Energy. The research team has been working closely with a number of companies, including EnerNex, EPRI, Riverside Public Utilities and Southern Company.