Not long ago, getting a virus was about the worst thing computer users could expect in terms of system vulnerability. But in our current age of hyper-connectedness and the emerging Internet of Things, when data is collected and transmitted at a rate and in quantities inconceivable a few decades ago, that’s no longer the case. With connectivity, a new law has emerged, one of universal concern to those who work in the area of systems control, like João Hespanha, professor in the departments of Electrical and Computer Engineering, and Mechanical Engineering at UC Santa Barbara. That law says, essentially, that the more complex and connected a system is, the more susceptible it is to disruptive cyber-attacks.
“It is about something much different than your regular computer virus,” Hespanha says. “It is more about cyber physical systems — systems in which computers are connected to physical elements. That could be robots, drones, smart appliances, or infrastructure systems such as those used to distribute energy and water.”
Much of Hespanha’s research is focused on developing algorithms to make sensor-rich, autonomous systems viable. And because those systems are both complex and connected — and, thus, vulnerable to attack — he also invests a great deal of time on security issues, or figuring out, as he says, “what happens if the sensors you're using have been compromised.”
“You have to work simultaneously on the two problems, designing a safe system, but also figuring out how to break the system by identifying its vulnerabilities and how somebody would attack it,” he adds.
The same holds true for the above-mentioned infrastructure elements, and in a paper titled “Distributed Estimation of Power System Oscillation Modes under Attacks on GPS Clocks,” published this month in the journal IEEE Transactions on Instrumentation and Measurement, Hespanha and co-author Yongqiang Wang (a former UCSB postdoctoral research and now a faculty member at Clemson University) suggest a new method for protecting the increasingly complex and connected power grid from attack.
The question that arises in any system that incorporates many sensors for monitoring is, what if someone intercepts the communication between two sensors that are trying to assess the health of the system? How does the system know not to believe — and act on — the false information?
Hespanha explains, “In the power grid, you have to be able to identify what the voltage and the current are at specific, highly precise points in time” for multiple points along the grid. Knowing the speed at which electricity moves, the distance between sensors, and the time it takes an oscillation to move between sensors, one can determine whether the oscillation is real.
Measuring voltage and current precisely has long been possible, but energy travels very fast, and to know exactly when a measurement is made at a sensor requires time resolution of just a few milliseconds. That used to be impossible to achieve, but something called a phase measurement unit (PMU) changed that.
PMUs, which are aligned with the atomic clocks used in GPS, can measure with extreme time resolution voltage and current anywhere in the grid. With the energy grid becoming increasingly distributed, and with photovoltaic cells, electric-vehicle charging, wind turbines, and other elements’ causing more perturbations, power providers now have to monitor the system more, and PMUs are among the most important devices for doing so. While PMUs could be used to inform autonomous control systems, so far, they have seen limited use, mostly for monitoring, for one simple reason: they are vulnerable to GPS spoofing attacks.
“There is the possibility,” Hespanha says, “that someone will hack the system and cause a catastrophic failure.”
The attack could be as simple as someone taking a readily available GPS jammer to a remote power-distribution station in the middle of the desert and tricking the system into providing false measurements. The net effect could be that the system, believing, for example, that there is a shortage of energy flowing into it because of the false readings, could overload power lines by dispatching energy production when and where it is not needed. That could lead to a cascade effect as the false readings ripple through the system, and incorrect actions are taken. Since it is virtually impossible to prevent someone from getting close enough to a remote substation to jam its GPS, Hespanha says, “What you need is a control system that can process the information to make good decisions. The system has to keep hypothesizing that what it is reading is not real.”
How It Can Work
"The power-supply system is a distributed system, so measurements are being made in many places," Hespanha explains. "If one of them starts to give erratic or unexpected measurements — a sudden current surge or a voltage drop — you should be able to determine whether those measurements make sense.”
In the case of an actual fluctuation, such as when many people in Los Angeles are using their air-conditioning on a hot summer day, the result may be a slight drop in the alternating-current frequency in the city. That drop creates a disturbance, which propagates along the entire western interconnect, which is the power grid stretching from western Canada south to Baja California in Mexico and reaching eastward over the Rockies to the Great Plains. As the disturbance travels through the grid, the power stations that feed the grid try to counteract it by generating extra power if the frequency is too low or decreasing production if the frequency becomes too high.
“You're going to start by seeing oscillation on the grid,” Hespanha explains. “That's exactly what the PMUs are looking for. You then compare the precise time you saw the disturbance in Los Angeles to the time you saw it in Bakersfield and then at other sensors as it continues north. And if those readings don’t reflect the physics of how electricity moves, that's an indication something’s wrong. The PMUs are there to see oscillations and to help dampen them to prevent them from developing."
But if someone fooled an automated system, instead of damping the oscillations, the PMUs could create them instead.
So how would such an attack be recognized and stopped? To illustrate, Hespanha draws an electrical line running between Los Angeles and Seattle, with many smaller, ancillary lines running off to the sides. “If power is going in a certain direction, you should also be able to see any oscillation in the side lines in that direction. And you know the physical model of what things should do, so an attacker who changed the measurement on the main line would also have to mess up a lot of other measurements on the side lines along the way. And that would be very difficult.”
Testing suggests that Hespanha’s system would be resistant to attack and remain effective even if one-third of the sensor nodes were compromised. “That would allow for a much more autonomous system; that’s the next big step,” Hespanha says. “This is an enabling technology that will be needed to make a lot of this control come online. And it will be needed soon, because the system gets more complex all the time and is therefore more susceptible to attack.”