California, August 2048. Much of the state has been baking in a severe heat wave going on for two weeks, with temperatures consistently reaching 100°F. The air is dry; the ground is parched; the flora is kindling. In this tinderbox, suddenly there’s a spark…
Within 15 minutes, the power grid’s artificial intelligence (AI) system detects signs of smoke and fire in satellite imaging and alerts the human operators on duty. They dispatch a drone to the area to confirm the nascent wildfire. The flying robot feeds live, high-fidelity video and imagery to the AI system to ascertain any potential risk to electrical infrastructure. According to the AI’s rapid assessment, the burgeoning blaze could threaten key substations and electrical poles, risking a blackout to tens of thousands of customers.
The AI then presents mitigation options to the grid operators. They quickly elect to reroute power in the area and bring an energy storage facility online, as well as a virtual power plant composed of thousands of customers’ electric vehicle and home batteries. Though a substation is soon lost, there is no blackout. Power continues to flow. The grid remains stable.
The Power Grid Today
America’s power grid is a complex, patchwork behemoth. As Department of Energy researchers Keith J. Benes, Joshua E. Porterfield, and Charles Yang described in an extensive 2024 report, “It consists of tens of thousands of power generators delivering electricity across more than 600,000 circuit miles of transmission lines, 70,000 substations, 5.5 million miles of distribution lines, and 180 million power poles. This system evolved organically over a century of piecemeal additions, and now operates at the heart of America’s $28 trillion economy.”
In this tangled web, aged wires, hodgepodge transformers, and rickety poles mesh with modern superconducting materials, smart grid meters, sensors, and ginormous lithium-ion battery installations. Somehow, it all works. The US power grid has an impressive overall reliability of 99.95%. And it has to be that good. American lives and jobs depend on 24/7 electricity.
“Even a power grid with 99% uptime would leave people and companies without power for 3.5 days in a year,” the authors note.
Now, our immensely intricate power grid is about to undergo perhaps its most disruptive makeover ever. To mitigate catastrophic climate change, operators must build and incorporate disparate, often intermittent, carbon-free electricity sources. Wind, solar, battery storage, hydropower, and nuclear are all needed in significant amounts. Moreover, thousands of miles of high-voltage wires must be erected to get the generated electricity to the places it is actually used. But is it possible to make these power systems work well together, while maintaining the grid’s impeccable reliability and reasonable cost? Artificial intelligence may be the key to making it all work.
The Control Room of the Future
The scenario at the beginning of this piece could not happen today, but it could play out sooner than one might think. In May of this year, a team of researchers from the National Renewable Energy Laboratory (NREL) published a technical report describing “eGridGPT,” their generative AI model “engineered to virtually support power grid control room operators by assisting in decision-making processes and interpreting the data and models.”
As the power grid grows in complexity — with increasingly variable power generation, electricity that flows both to and from customers, and heaps of data from widespread sensors — the NREL scientists imagine that it may become too much for human grid operators to comfortably manage on their own. So they developed an AI assistant “to act as an interface between a screen in front of the operator and the orchestrator for the comprehensive processing of large volumes of data, scenarios, and digital twin simulations.”
In a future grid, the NREL researchers see humans as the decision-makers, but AI will inform those decisions and potentially carry them out.
Seong Choi is the engineer lead at NREL’s Power Systems Engineering Center and lead author of the eGridGPT report.
“The primary objective of eGridGPT is to assist operators in near real-time decision-making by analyzing large datasets, recognizing patterns, simulating scenarios, and proposing mitigation strategies,” he explained to Freethink.
Choi says that some grid operators have already demonstrated interest in testing it out.
AI and the Power Supply
With the rise of large language models and the electricity-hungry data centers required to power them, AI has recently been viewed as an impediment to decarbonizing the grid. But many scientists and grid experts say that AI systems utilizing deep learning and machine learning are exactly what’s needed for a futuristic grid powered by renewables.
“By harnessing the capabilities of AI, it is possible to develop intelligent systems that can adapt to dynamic environmental conditions, forecast energy production, and optimize resource allocation,” a team of engineers from the University of Johannesburg in South Africa wrote in a review published earlier this year.
The model could potentially cut total maintenance costs by half.
Incorporating AI into the power grid could bring about a host of benefits. For starters, it can monitor reams of data about individual assets like solar panels, wind turbines, and inverters to identify what needs maintenance, thus reducing downtime, maximizing output, and reducing costs for utilities and ratepayers.
Energy equipment giant GE Vernova already utilizes AI for predictive maintenance on wind farms. By analyzing data from sensors on wind turbines, machine-learning algorithms predict potential equipment breakdowns before they happen. Scientists at Argonne National Laboratory have developed a failure-predicting AI model that works for a range of components, allowing utilities to schedule maintenance accordingly. In one case where they tested it on solar inverters, they found that the model could potentially cut total maintenance costs by half and unnecessary crew visits by two-thirds.
Another potential benefit of AI: more closely matching supply with demand. To ensure electricity availability, the former must always match the latter. But this becomes more difficult as more electricity comes from intermittent wind and solar vs. always-on power plants, like nuclear, or natural gas plants that can be fired up on demand. AI can better forecast cloud cover for solar panels and wind patterns for turbines, helping predict future supply. In turn, it can also use utility data to better predict what demand might be throughout those days. With this data, grid operators can decide if they need to pull more resources from other regions or activate battery-storage assets.
“As more renewable energy sources come online and extreme weather events become more frequent, operators face uncharted territory,” Choi adds. “They often rely on informed guesses to address these new challenges, while AI can offer projections for potential future scenarios.”
On the ground level, AI models can also help optimize the positioning of solar panels to capture the most sunlight throughout the day. For wind turbines, it can adjust blade angles to optimize energy capture. With batteries, it can monitor market conditions on the fly, discharging energy when it’s expensive and storing it when it’s cheap. Tesla, one of the leaders in battery storage, already utilizes AI algorithms to do this.
Drawbacks
Ultimately, the power grid is the sort of data-sensitive machine that lends itself perfectly to artificially intelligent automation. But there are a few reasons why it may prove difficult to integrate AI.
First among them is security. When control of the grid is increasingly in the hands of software, it becomes more susceptible to traditional cyberattacks. But machine learning models are also at risk of a more specific form of subterfuge: data poisoning. If these systems are somehow fed garbage data, they could make decisions that result in mass power outages.
Accuracy is another issue. There’s concern that these models could, over time, grow biased in ways that misalign their goals with the fundamental need to ensure electricity access to all.
“Accuracy remains a significant challenge, particularly in building operators’ trust. Building this trust is a critical goal for AI,” Choi told Freethink. “AI relies on historical data for learning and lacks awareness of current conditions, which means it may not always provide up-to-date insights.”
Second, AI is not without costs. As the University of Johannesburg researchers noted, “While AI can optimize renewable energy systems and reduce carbon emissions, it also comes with its own environmental footprint. Training AI models requires significant computational power, resulting in high energy consumption and associated carbon emissions.”
Essentially, the grid must evolve into one giant computer.
Third, and most daunting, to get to a point where AI can make reliable decisions on the grid-scale, the grid itself would require massive upgrades.
“While today’s grid primarily moves electricity in one direction from large, centralized power plants to electricity customers with relatively little information exchange, the grid of the future will manage multi-directional flows of energy and information across a diverse set of grid-connected resources,” the DoE researchers wrote.
Sensors galore will need to be installed across the network to feed reams of data into the models. SCADA (supervisory control and data acquisition) systems comprising a range of electrical components will need to be inserted across grid assets. Smart meters will need to become ubiquitous in homes and businesses. Essentially, the grid must evolve into one giant computer. Completing this unprecedented transformation will cost hundreds of billions of dollars.
Worth the Cost
Yet it will likely be worth it in the long run. The global cost of climate climate change, encompassing damages to infrastructure, property, agriculture, and human health, is projected to tally between $1.7 trillion and $3.1 trillion per year by 2050. A colossal buildout of renewable energy is needed to avert those costs, and if AI ultimately makes this carbon-free grid cheaper and more efficient, so much the better for people and for the planet.
We’d love to hear from you! If you have a comment about this article or if you have a tip for a future Freethink story, please email us at tips@freethink.com.
