Like a living organism that evolves in order to adapt to the changing demands of its environment, the labyrinthine US electric grid comprises more than 9,200 generating units with over a million megawatts of generating capacity connected to a 300,000-mile network of transmission lines.

This marvel of engineering, built in the 1890s, is now struggling to cope with the groundswell of complex digital and computerised equipment and technology dependent on it. Federal data released by the Department of Energy (DOE) and the North American Electric Reliability Corp (NERC) reveals that the US endures more power blackouts than any other developed nation. The electric grid loses power 285% more often than in 1984, when records began, at an annual cost to businesses of $150bn. The root causes include ageing infrastructure, a lack of investment in grid modernisation, and policy gaps at local and federal level, exacerbated by increased energy demand.

The US Government set aside $4.5bn for smart grid development as part of the American Recovery and Reinvestment Act of 2009 in order to create a robust, interactive 21st century power system capable of withstanding multiple threats including earthquakes, solar flares and terrorist attacks.

"The grid is overburdened and suffering more faults than ever, and migrating toward the smart grid is widely recognised as the future."

Remotely detecting and locating faults quickly and accurately to prevent outages and equipment damage is a fundamental element of that overarching strategy and a research team at Binghamton University in New York believes its singular spectrum analysis (SSA) algorithm may provide the answer.

“Big data and Internet of Things (IoT) technologies are leading us into the era of the smart city and the smart space, but so far the ageing US power grid system is not seeing much benefit from these advancements,” says Yu Chen, associate professor of electrical and computer engineering, who leads the Binghamton team comprising Zekun Yang, a PhD student in electrical and computer engineering, Ning Zhou, assistant professor in electrical and computer engineering, and Aleksey Polunchenko, assistant professor in mathematical sciences.

“The grid is overburdened and suffering more faults than ever, and migrating toward the smart grid is widely recognised as the future. Situational awareness is essential as it help operators to respond to major events − a disturbance in a power grid, for example – in a timely and efficient manner.

“However, current solutions are not able to catch the anomalies quickly and with sufficient accuracy. In 2015, the Binghamton University team came up with the idea that anomaly detection could be addressed as a quickest change-point detection problem, which involves the design and analysis of procedures for on-the-go detection of changes in the characteristics of a running (random) process. Our goal was to develop a detection mechanism that is faster and more robust in the noisy environments.”

New and improved: the EST algorithm and the Power System Toolbox

Electric grids may be vast, but the principle behind them is still relatively simple. From a generator, through wires, a light bulb and back again, there can’t be a single break in a circuit or nothing works. Multiple complete circuits − a grid − keep electricity flowing even when one wire does go down.

This redundancy provides stability but is filled with vulnerable points. Beyond a tree limb taking out a wire in a windstorm, hackers can break in and subtly change how electricity flows, which can have a potentially catastrophic effect on banking, communications, traffic and security infrastructure.

At present, the time and location of anomalies in the power grid is determined by formulas such as the event start time (EST) algorithm, which calculates differing arrival times of power changes in different geographic locations. Even though the differences are incredibly small, they are enough to triangulate the location of changes, allowing authorities to pinpoint where an outage has occurred.

The Binghamton team used simulation data from the Power System Toolbox (PST) – the software used for power system computation, analysis, and control − to prove that the SSA algorithm is faster and more robust at detecting changes in the power grid from generator or transmission line problems.

“The major challenge lies in parameter setting to balance the trade-offs between computing delays and accuracy,” Chen explains. “The PST generates simulative datasets that describe the behaviours of a 16-machine, 68-bus system when faults happen at either generators or on transmission lines. Variant scenarios were then generated corresponding to different disturbance characteristics.

“The SSA theory has demonstrated its ability in a wide range of practical applications, such as finding data structure, extracting periodic pattern and complex trends, and smoothing and detecting change points. For example, researchers applied SSA to forecast accidental monthly deaths in the US, concluding that it offered a more accurate result than other series data analysis methods.

“Through extensive experimental study, we have validated the effectiveness and correctness of our SSA algorithm − and it outperformed the well-known EST algorithm completely,” he adds.

Real-world scenario: how the SSA detects grid anomalies

Chen and his four-strong team adapted and improved the SSA algorithm to improve its effectiveness at detecting grid anomalies, underlining its superiority compared with the well-known EST algorithm.

"Theoretically, the SSA approach features lower false alarm rates and higher detection accuracy."

“Theoretically, the SSA approach features lower false alarm rates and higher detection accuracy; on a practical level, our matrix operation-based SSA conversion is more regular and easier to implement than hash-function-based cardinality estimator. The SSA procedure results have already outlined the potential point of change, which simplifies follow-up operations and benefits the overall process.

“This character is unbeatable by any known approach,” he continues. “The experimental results at Binghamton have demonstrated that our SSA algorithm is not only faster and more robust in noisy environments, but also is able to capture more subtle disturbance that the existing EST algorithm could not.”

In order to illustrate how the SSA solution could be used in the real world, Chen describes a potential scenario whereby the algorithm is employed to detect disruptions in the power network.

“In the smart grid, surveillance sensors are embedded in power networks in order to monitor the operation and health of the grid,” he explains. “The SSA algorithm can be built in every sensor so that the electromechanical signals travelling through the sensor are scanned and analysed by SSA. When any abnormal change is detected, the sensor raises alert to the central control unit. Based on the times and locations of the alerts, the central control unit is able to quickly identify the source of the anomaly, which can take the form of an accident or a malicious attack on the electricity grid.”

Detection vs prediction: the future of network monitoring

Despite the excitement surrounding the breakthrough, the approach still requires fine tuning, including more ways to gather accurate geolocations of problems, more simulation testing and real-world data collection in order to validate the algorithm and polish it to cope with more realistic scenarios.

“We have confirmed the SSA's effectiveness, but some further improvements are needed in order to take it out of the laboratory and into the marketplace,” Chen confirms. “At the current stage, the algorithm can only detect and locate problems. It is a solid foundation for the next step: prediction.

“Being able to detect subtle changes in the power grid promptly, our approach is very promising to predict future problems by integrating measurement data with a power system model.”