MIT Technology Review learned that California’s statewide grid operator is set to become first in North America to deploy AI to manage outages.
We wanted to modernize grid operations. This is a perfect fit for that,” Gopakumar Gopinathan says, a senior adviser on power system technology at the California Independent System Operator (CAISO), pronounced KAI. “AI has already transformed different industries.” But we haven’t seen many examples being used in our sector.
On July 15, CAISO will announce a deal with OATI to run a pilot using their new AI software, Genie. The software uses generative AI for grid operators to analyze and perform real-time analyses. It also has the potential to make autonomous decisions about key grid functions, similar to the switch from uniformed police officers to stoplights with sensors.
While CAISO delivers electrons to cutting edge Silicon Valley companies and labs, the actual task managing the state’s electric system is surprisingly analogue. CAISO engineers today scan outage reports to find keywords that indicate planned or ongoing maintenance, read the notes and then load the items into the grid software program to run calculations about how a downed transformer or line might affect power supply.
Abhimanyu Takur, OATI Vice President of Platforms, Visualization, and Analytics, says that even if scanning one outage takes less than a moment, multiplying it by 200 or 300 outages adds up. “Then, different departments do it for their respective keywords. We can now consolidate this into a single dictionary and AI will scan it and generate the report.
Gopinathan says that if CAISO finds Genie to be more reliable and efficient in managing outages then the operator might consider automating other functions on the grid. “I think after a few testing rounds, we’ll know when to call it a success or not,” he says.
The experiment, regardless of its outcome, marks a significant change. Richard Doying, a former top executive of the Midcontinent Independent System Operator (the grid operator for 15 states from upper Midwest to Louisiana), says that most grid operators use the same systems utilities have been using “for decades”.
Doying, a vice-president at Grid Strategies, says that these organizations are divided into teams of people who work on very specific, specialized jobs and use their own proprietary tools they’ve developed. “To the extent some of these new AI-based tools can draw data from different areas of an organisation and conduct more sophisticated analyses, that’s only beneficial for grid operators.”
Report found AI could speed up studies on grid transmission and capacity, improve weather forecasting so that it can predict how much energy solar and wind plants will produce at a certain time, and optimize the planning of electric-vehicle charger networks. A report from the energy department’s Loan Programs Office concluded adding more “advanced technology” such as sensors to different pieces of equipment would generate data that could enable AI to do more over time.
The PJM Interconnection, the nation’s largest grid system spanning 13 states in the densely populated mid Atlantic and Eastern Seaboard, took a major step towards AI in April by signing a deal to use Google’s Tapestry software for improving regional planning and accelerating grid connections for new generators.
ERCOT (the Texas grid system) is considering adopting similar technology to what CAISO will be using, according to an anonymous source who was familiar with the plans and not authorized to speak in public. ERCOT declined to comment on a request.
Australia provides an example of how the future could look. In New South Wales where grid sensors and smart technologies are more widely used, AI software rolled out in Feburary is now predicting and adjusting the amount of power that rooftop solar panels can provide to the grid.
Until recently, the majority of the AI and energy discussion has been focused on the power demands of AI data centres (for more information on this, check out MIT Technology Review’sPower Hungry series).
Charles Hua, coauthor of a report from the Energy Department last year, and executive director of PowerLines (a nonprofit that advocates improving the affordability and reliability US grids), says, “We have been talking about what the grid can be used for AI, but not nearly enough about what AI can be used for the grid.” “In general there’s an enormous opportunity for grid operators and regulators to use AI effectively, and harness it for more resilient, modernized and strengthened grids.”
Gopinathan says he remains cautiously optimistic for now. “I don’t like to overhype this,” he says.
He adds that “it’s still a first step towards bigger automation.”
Right now, it is limited to our outage-management system. He says that Genie hasn’t yet been able to communicate with other parts of the company. “But I can see a future where AI agents will be able to do much more.”
