The U.S. power grid is like the body’s central circulatory system — it provides the energy without which virtually nothing functions. In the past, it was a self-contained system, with few openings to the outside world. Now, however, the systems for managing it have become much more sophisticated: optimizing distribution as demand changes, preventing blackouts, integrating a variety of energy sources, and more. Artificial intelligence (AI) is becoming a major tool in these grid-management systems.
As systems get more complex, however, the chance of unintended consequences multiplies. As part of the design process, it is therefore vital to consider the possibilities of what can go wrong. AI is a powerful tool for improving grid management, but its increasing use opens up possible sources of trouble ranging from unpredicted events to intentional hostile interference.
“The rapid proliferation of AI promises significant value for industry, consumers, and broader society, but as with many technologies, new risks from these advancements in AI must be managed to realize its full potential.”1
As with most things, there’s an upside and a downside to the relationship between AI and cybersecurity.
Advantages
Incorporating renewable energy sources such as solar and wind into the energy grids brings many rewards. Not only are they more economical and less polluting than fossil fuel-burning power generators, but they have the potential to help stabilize the electricity supply. For example, they can be quickly switched on to provide more power in the event of a surge in demand or switched off when the demand drops. This is a great improvement over using peaking power plants that must be started up and brought online to meet a demand surge.
Another advantage of renewables is that since solar and wind facilities are widely distributed geographically, each installation provides a smaller percentage of the overall energy than a large power plant. The loss to the overall grid system will therefore be reduced in the event of a localized problem like storm, flood, or fire.
This decentralization, however, requires complex systems of information and control to manage the interconnection of distributed energy resources (DERs). For example, the energy from solar reaches a peak during midday, but wind, depending on where the turbines are located, may or may not be fairly steady around the clock. AI can help by using data about current and predicted patterns of sun and wind for any particular location to decide the best way to stabilize the grid.
AI is also a valuable tool for coordinating the different types of energy sources, deciding which ones should be brought online to meet increased demand and which should be switched off when the availabile capacity is greater than the demand.
AI can even detect patterns of behaviors on the grid that warn of impending trouble so they can be addressed before they cause outages.
Risks and Solutions
AI is indispensable for managing all this complexity, but at the same time it opens new windows for potential trouble. Paradoxically, however, the vulnerabilities opened up by using AI on the grid can also be combatted by using AI.
“Because the traditional utility network was physically isolated from the public network, an IT approach was sufficient for most threats,” said Bo Chen, an Argonne National Laboratory computational engineer. “Today’s utility network creates more vulnerabilities as new technologies are integrated. Many sophisticated attacks can hide themselves so an IT approach cannot detect them.”2
Several institutions are partnering with the Department of Energy’s Oak Ridge National Laboratory (ORNL) to launch a project to develop an innovative suite of tools that will employ machine learning algorithms for more effective cybersecurity analysis of
the U.S. power grid. The suite, called AI-PhyX, is designed to “streamline collection and analysis of data to comprehensively tackle all facets of cyber resilience, including vulnerability analysis, attack detection, threat mitigation, and system recovery. AI helps convert data into actionable information that enables system operators to make better decisions.”
The effectiveness of AI-PhyX is that it integrates different cybersecurity applications to take the best parts of all of them and provide the most coordinated, most effective, AI protection against as many different kinds of threats as possible. The suite will also include analysis tools to help identify particular vulnerabilities in the interconnections of different DERs. A goal of these defensive techniques is that even if there is a security breach, the grid will continue to function.
Conclusion
It is vital that cybersecurity has to be integral to the design of any AI system, but never more so than when the system is the source of energy for all other systems and devices.
