In a report released this week, the Department of Energy outlined a detailed view on where and how artificial intelligence could be integrated to modernize the grid — at least in the near term.
- The top line: In the sweeping report, DOE outlined priority use cases and suggested guardrail requirements for deploying artificial intelligence for the energy transition more broadly, and avoiding pitfalls. For the grid, it focuses on the potential of foundation AI models — those trained on vast amounts of data, like OpenAI’s GPT — to improve grid planning, permitting, operations, and resilience.
- The market grounding: Despite the power sector’s ongoing trend toward digitization and push to collect more data, utilities in particular still say they lack a central, holistic vision of AI deployment. Some utilities are experimenting with AI’s potential, but they tend to be small-scale programs, and leaders at even the most innovative utilities predict that meaningful applications of AI for the grid may still be a few years out.
- The current take: The report described a “confluence of change” in the power system — no doubt including the huge load growth predicted to come from data centers, electrification, and domestic manufacturing — that is creating a need for “massive innovation” in the grid especially. “This innovation will be driven by technology that can transform vast amounts of information into actionable, decision-making capabilities. Huge advances in AI in recent years present an unparalleled opportunity to accelerate innovation across every aspect of the clean energy transition.”
The report essentially comprises a laundry list of high-impact potential applications for AI — but comes with a note of caution.
Effectively integrating AI into the power industry will require stringent guardrails, DOE said; and that’s particularly true for the use of foundation models within the power industry, given the potential for bias and extrapolation, and cybersecurity vulnerabilities, among other challenges.
“The electrical grid of the United States is among the most complex machines on Earth,” the report found.
The department outlined a handful of requirements for the safe application of any type of AI: 1) systems should be“rigorously validated,” 2) take “human-in-the-loop” approaches; 3) outputs should be “physics-informed and explainable;” 4) applications should be scalable; and 5) they should comply with “robust cybersecurity policies and standards.”
Matt Casey, who leads the Latitude Intelligence research team, said that while the security risks and concerns around data quality, gaps, and bias are well-known, there are already real world AI deployments in the field today that utilities and other industry stakeholders (like regulators and policy-makers) can use as blueprints and proofs of concept when developing frameworks and strategies for higher risk, more critical workloads. (Casey’s team recently published a report assessing the current “state of play” of utilities’ use of AI.)
For most AI use cases, both on the grid and elsewhere in the energy transition, Casey added, the two most key factors driving progress have been better access to new data, coupled with AI’s ability to analyze multiple disparate data sets.
AI for grid planning
Grid planning is “ripe for AI-enabled innovation,” DOE said, thanks to the fact that planners are already grappling with a surge of data from things like smart sensors and satellites.
DOE’s deep dive into AI for the grid comes just two weeks after the release of its latest “liftoff report” focusing on pathways to deploying grid-enhancing technologies and other innovations at scale. Given the need to vastly increase the capacity of the grid, and to do so in a time period considerably shorter than most new transmission builds, the liftoff report outlines a suite of tools that can be deployed from the grid planning arsenal to avoid building new fossil fuel generation. This additional research on AI essentially adds another technology to the department’s grid-planning arsenal.
Integrating AI will require some significant adaptations by regulators, who will need to incentivize utilities to leverage the technology to optimize existing assets — rather than building the new capital projects that are proving increasingly hard to get interconnected. Transmission operators and utilities, for their part, will need to both digitize operations and break down data silos while prioritizing customer privacy and infrastructure security, DOE noted.
The electrical grid of the United States is among the most complex machines on Earth.
Assuming those challenges can be balanced, AI can help the power industry move away from the knee-jerk reaction of like-for-like infrastructure replacement projects in light of load growth — often of gas plants, which tend to be easily justified and incorporated into the rate-base — by increasing system knowledge and awareness.
For instance, completing, correcting, and harmonizing “sparse” data on grid infrastructure, as well as validating existing datasets, “would allow utilities to employ predictive asset replacement to leverage economies of scale that are not realized through the existing maintenance paradigm,” the report said.
AI can also enhance capacity modeling, thereby boosting adoption of grid-enhancing technologies like dynamic line rating and topology optimization. AI can support DLR through “improved forecasting of local and regional weather” and topology optimization by identifying opportunities for deployment more quickly and cheaply.
And the tech can also have an impact on the pesky interconnection process, which remains a major bottleneck in the energy transition, by accelerating grid modeling and application compliance processing. For instance, large language models could help to screen and validate unstructured data in those applications, such as documents pertaining to land ownership.
AI for siting and permitting
Meanwhile, Casey said that siting and permitting processes are “low hanging fruit” for immediate and impactful AI utilization.
“There is an abundance of unstructured, siloed data [involved in the process] that when combined, synthesized, and analyzed can provide immediate benefits by helping expedite some of the most time-consuming steps in project development,” he added.
As the report pointed out, permitting timelines are a “key sensitivity driving deployment cost for geothermal energy, distributed rooftop solar, and critical minerals mining.”
However, AI can simplify the onerous work involved in both creating and processing these applications. LLMs can extract and organize unstructured data like past permits, approvals, and environmental reviews, the report said, pulling that information into structured data that can then be deployed to fine-tune models for specific contexts, building tools for everyone from developers and government reviewers to the public.
They can also assist with processing public comments, streamlining the application process, and training less experienced workers in the siting and permitting ecosystem.
Just last week, DOE issued final rules to streamline transmission permitting to the extent of the department’s ability, including via placing a two-year deadline for the federal permitting process. That rulemaking, however, will not impact the state-level process, which this week’s report highlighted as an additional area where AI can make an impact.
AI for grid operations and reliability
There’s also an opportunity for AI to help grid operators better maintain assets, forecast generation, make use of flexible loads, and inform operational safety, DOE found. However, grid operations are “highly security-sensitive” applications for AI.
“Overcoming this obstacle necessitates collaborative efforts between academia, government, and industry to establish data-sharing mechanisms, fostering an environment conducive to advancing R&D — while still ensuring that necessary and appropriate security and privacy considerations are upheld,” the report found.
That might mean the creation of “shared data warehouse infrastructure” and “common performance benchmarks,” the department suggested. Those should be built using “synthetic privacy-preserving data modeled after real-world data” — in other words, realistic but fake data — and shared with “trusted researchers.”
AI-powered models designed to mimic large and complex systems like the power grid can greatly speed up traditional simulation techniques. For example, the technology could help predict weather conditions with significantly more detail, which is key for accurately forecasting renewables production.
That’s not just the case for centralized grid operators, the report said, but also for virtual power plant providers.
“AI can help optimize revenue for VPP providers while also providing customized options based on user preferences,” the report said. (Smoothing out the data-sharing pathways would also eliminate a “key bottleneck” for VPP deployment.)
DOE also pointed to the potential for predictive maintenance to allow for the phasing out of schedule-based approaches to asset servicing, leading to “significant cost reductions.”
Anecdotally, Casey said, grid operations is the use case where there’s been particularly robust activity to date. The concerns around grid reliability — driven largely by load growth, demand growth, and the emergence of DERs — are well-known and well-documented. And Casey added that they have driven both utilities and vendors to prioritize developing new solutions to address them, such as “new” data sources like sensors and drones. That new data, combined with historical supply and demand data and weather data, is already helping improve and secure grid operations.
Specifically, Casey said Latitude Intelligence has seen particular activity in asset inspection, infrastructure condition monitoring, field maintenance management, wildfire mitigation and response, and vegetation management.
AI for grid resilience
Recent grid operations gains notwithstanding, Casey said resilience and long-term planning are where the power sector has the most to gain from AI.
“The number of variables utilities (and the sector in general) need to evaluate when long-term planning is only growing and becoming more complex,” he said.
The DOE report found that AI’s potential to “rapidly invest vast amounts of data and identify subtle shifts in data distribution” is therefore extremely powerful. The department laid out a framework for how three different types of machine learning — natural language processing, image pattern recognition, and inference — can serve grid resilience functions.
NLP, for example, could be used to support operator and dispatcher decision making, and identify and localize faults based on SCADA streams. Image pattern recognition can be used for outage prediction and predictive asset maintenance, through the use of sensors and data analytics; meanwhile, cause and effect forecasting can be used to detect anomalies in asset data, make inferences in environments with sparse data by filling in the gaps, and quantify the uncertainty in model predictions.
Beyond the grid
Finally, the report outlines a broad swath of use cases elsewhere in the clean energy economy. Included on the long list are AI’s potential to optimize planning for EV charging networks, conduct quality assurance testing in manufacturing, and discover new materials.
However, actually implementing AI on the power grid faces several hurdles beyond utility adoption and regulatory structure. For instance, the report points to the technology’s significant energy use (which is of course one of the driving factors behind predictions of major load growth), immense workforce needs, and the potential for AI’s widespread use to have disproportionate negative impacts on certain communities.
Modernizing and decarbonizing the grid are essential components of the country’s ability to meet the Biden administration’s clean energy goals, the report said. That means building a grid that “allows grid managers to make decisions based on multi-directional flows of energy and integration,” quickly integrates new clean generation sources, “actively balances both electricity supply and demand,” and mitigates changes and challenges from climate change.
That is an undoubtedly tall order, one that comes with both “great opportunities and potential risks,” DOE said.
