Utility companies are increasingly finding uses for artificial intelligence (A.I.) to enhance the electric grid’s resilience, efficiency and performance—and the future looks even brighter.

Preventive maintenance

Utilities are leveraging A.I. to understand “literally decades of precursors” to an equipment failure so they can repair or replace a piece of equipment before it fails, said Scott Aaronson, senior vice president of energy security and industry for the Edison Electric Institute, Washington, D.C.

“All of this is adding reliability, resilience and efficiencies to the way that electric companies operate the grid,” Aaronson said.

A.I. algorithms analyze sensor data to monitor asset health and detect early signs of failure in transformers, lines and breakers, enhancing operational reliability and reducing maintenance costs, according to Khalid A. Behairy, managing director at Deloitte, New York.

A.I.-enabled drones with advanced image processing capabilities allow electric companies to proactively maintain and upgrade assets during routine inspections by identifying and forecasting potential failures and outages, Behairy said.

Southern California Edison (SCE) uses A.I. and predictive analytics to predict asset failures and prioritize maintenance, which helps cut costs, said Jeff Monford, senior advisor for corporate communications.

“A.I. learns patterns from load, heat and weather to predict failures,” Monford said. “Also, SCE has built a model to predict the probability of failure for overhead and auto switches. By using the ‘YOLO’ [you only look once] computer vision model, SCE identified the correct types of switches from inspection photos, improving the accuracy of the switch failure model.”

Minimizing outages

A.I. is widely used in asset monitoring and diagnostics, assessing asset performance and understanding potential for failure and degradation in performance, said Aidan Tuohy, director of power systems, transmission operations and planning at the Electric Power Research Institute (EPRI), an independent, nonprofit energy research and development organization based in Washington, D.C.

The use of A.I. is also improving situational awareness, Tuohy said. A.I. algorithms process vast streams of synchrophasor data, also referred to as phasor measurement units, data from supervisory control and data acquisition (SCADA) networks, and other sensor inputs to detect patterns that may indicate impending grid instability or equipment failure. This provides operators with a more accurate and forward-looking view of grid health.

A.I. also enables faster outage detection and restoration, Behairy said. A.I. systems can detect faults within seconds—much faster than traditional systems, pinpointing outage locations with meter-level accuracy and helping crews restore power more quickly.

SCE’s AWARE tool is an advanced solution for identifying faults, their locations and types by using high-fidelity waveform recordings, advanced metering infrastructure information, grid models and SCADA data in real-time, Monford said.

“It uses machine learning to differentiate between various disruptions, such as objects hitting a wire, underground issues or transformer failures,” he said. “This capability increases system reliability and allows for immediate detection and response to potential disruptions. The tool is already being used in several locations and has been recognized for its innovation by the Edison Electric Institute.”

Vegetation and wildfire risk management

SCE leverages A.I.-powered remote sensing tools, including LiDAR and satellite imagery, to monitor vegetation encroachment and wildfire risks, Monford said. These tools reduce the need for manual inspections and enable predictive analysis to anticipate threats before they materialize.

Indeed, vegetation encroachment is a top cause of outages and wildfires, Behairy said. A.I. enables utilities to detect vegetation overgrowth using LiDAR, satellite and drone imagery; predict tree fall risk based on weather and terrain data; and optimize trimming routes and crew schedules—improving efficiency and risk mitigation. One utility cut fire ignition by 72% in high-risk areas using this approach.

“Computer vision, powered by pretrained image processing models, enables drones to detect anomalies such as heat and vegetation encroachments and upload real-time data during flights,” Behairy said. “The models also continue to improve inspection data accuracy.”

For wildfire risk resilience, utilities are taking A.I.-enabled cameras and data from weather sensors from all over their systems to understand where fire risk is highest, Aaronson said. They are also using A.I.-enabled cameras and satellite imagery using the color of the topography and trees to identify where fire risk is greatest.

“They then can more effectively reduce risk where a fire may occur for any reason, whether it’s a lightning strike or an unattended campfire or something else,” he said. “They are able to see smoke using A.I.-enabled cameras and then can work with first responders to deploy as quickly as possible and knock those fires down.”

Utilities are partnering with the Department of Energy and National Labs to determine even better targeted ways to identify where fire risk is, make risk-informed decisions about operations and deploy first responders, Aaronson said.

Forecasting demand

A.I. has been used for forecasting demand, renewable energy production and other variables for many years, leveraging traditional A.I. techniques such as machine learning, Tuohy said.

“More recently, new methods such as A.I. weather forecasting are being adopted,” he said. “This allows for operators to be better prepared for events that might happen on the grid, as well as efficiently schedule resources to meet demand.”

A.I. helps manage distributed energy resources such as solar, storage and electric vehicles by balancing real-time inputs with load requirements, regulating voltage and frequency from inverter-­based systems, and enabling dynamic, decentralized control at the grid edge to ensure stability and efficiency, Behairy said.

Additional uses

A.I. can help secure grid infrastructure by detecting anomalous behavior in SCADA and OT networks, identifying threats across smart meter and distributed energy resources (DER) communication channels, and enabling real-time, automated incident response. PJM Interconnection integrates A.I. tools to enhance cyber resilience, Behairy said.

Then there are the next-generation smart meters, he said. AMI 2.0 devices use A.I. and GPU-enabled capabilities at the grid edge to make real-time decisions and better integrate two-way power flow.

“Additionally, AMI 2.0 will empower customers with more timely and accurate information, enabling better energy management, faster outage restoration and new service offerings targeting DER value realization,” Behairy said. “A.I. analyzes smart meter and behavioral data to create personalized customer profiles, enabling utilities to offer targeted energy efficiency programs, or time-of-use rate plans tailored to specific usage patterns.”

Future A.I. applications

Future A.I. systems will detect faults and autonomously reconfigure grid topology in response—creating self-healing grids that minimize outages and protect infrastructure without human intervention, Behairy said.

“A.I. will also help DERs participate dynamically in energy markets, reacting to price signals, grid stress or weather forecasts,” he said. “This turns passive assets into grid-interactive resources, supporting reliability while optimizing for cost and emissions.”

Moreover, the Department of Energy has funded research using A.I. to accelerate the development of new circuit technologies for high-voltage DC converters to help more efficiently integrate remote renewable generation, Behairy said.

In the future, A.I. will increasingly be leveraged for grid planning, Monford said. SCE is piloting physics-informed A.I. powerflow models in collaboration with ThinkLabs and NVidia to automate grid energization requests and accelerate interconnection of customer loads, improving reliability and planning accuracy.

“Use of A.I. will be a key tool in achieving California’s decarbonization goals for 2045,” he said. “Success depends on robust data governance, cross-functional collaboration and continued investment in secure, scalable infrastructure. As the grid becomes more dynamic, A.I. will be essential for managing complexity, ensuring reliability and delivering customer value.”

Equally important is a forward-looking workforce strategy, Monford said. SCE is actively evolving its talent model to meet the demands of a digital, A.I.-enabled grid—reassessing skills, updating recruiting methods and forming university partnerships to embed A.I. into power engineering curricula.

A.I. is increasingly being used to speed up the generation interconnection process by compiling large amounts of information from various forms and data to streamline and accelerate processes, Tuohy said.

“EPRI has developed tools to screen future operating conditions so planners can assess the most stressed operating conditions that may be faced,” he said. “For example, screening large amounts of data about future wind and solar production, generator dispatch, load and transmission flows to determine the situations when the grid may be most congested, leveraging machine learning and other A.I. applications.”

EPRI has also been investigating the ability of publicly available large language models (LLM) to perform various power system assessments, Tuohy said. The research has shown that such tools, with input and iteration from subject matter experts and with the appropriate data, can be used to support system protection functions.

EPRI is extending this concept to other transmission planning functions to develop a planning orchestrator leveraging agentic A.I., allowing the LLMs to perform increasingly complex power system simulation studies that may not presently be achievable by a single power engineer with existing simulation tools, he said.

“A.I. applications hold significant promise to improve reliability, increase efficiency and meet industry challenges,” Tuohy said. “However, it is crucial to get the underlying systems and capabilities right before widespread deployment and benefits can be seen. This includes data governance, cybersecurity and staff training, all of which need to be priorities as A.I. tools are used more widely.” 

As A.I. technology develops, it’s very exciting to see how electrical companies will best use it going forward, Aaronson said. Still, utilities and policymakers need to go into this new era of technology with an eye toward ensuring security and resilience of this most critical infrastructure. For all those efficiencies, there are also potential malicious uses of A.I. that need to be defended.

A.I.’s limitations also need to be understood, he said. For example, “hallucinations” are increasing—conclusions formulated by A.I. with inaccurate or erroneous data.

“If the data is not good, the output won’t be good, so we need to be very meticulous about both the security of the infrastructure and the trustworthiness of the data. This is why we need that human in the loop to check the work of A.I., especially as we think about truly vital applications like critical infrastructure operations,” Aaronson said.

STOCK.ADOBE.COM/AREE, STOCK.ADOBE.COM/KOSSSMOSSS