EPRI Leads Open Power AI Alliance: Data Sharing Drives Grid Modernization

The combination of power system modernization and artificial intelligence (AI) is being hindered by a lack of system data available for AI training—a judgment widely shared by experts in the technology and power industry.

"Data is to AI what oil, natural gas, wind, and solar are combined." Alexandr Wang, founder and CEO of Scale AI, a leading technology market company, testified before the U.S. House Energy and Commerce Committee on April 9, "Every major AI advancement is rooted in data shaped by human expertise," and "to win in AI, you must turn data assets into an advantage."

Power companies are equally eager to win with AI, a consensus formed among power system executives and analysts.

"The grid is one of the most complex machines on Earth," and "it is ripe for digitalization," noted Shinjini Menon, senior vice president of system planning and engineering at Southern California Edison. She added that the data underpinning digitalization is critical because "there is not enough time, money, or resources on Earth" to continue with past practices.

To break down data accessibility barriers, the power and technology industries have joined forces to advance the "Open Power AI Consortium," led by the Electric Power Research Institute (EPRI). Meanwhile, regulation and technological innovation are opening the door to data access. But experts told Utility Dive that serious disagreements over cybersecurity and power customer rights remain to be resolved.

Why data matters and where it comes from

Many stakeholders view new AI capabilities as a tool to address the vast scope and high complexity of the power system.

"This is not your grandfather's AI—it is not just machine learning, but it still needs data to learn," said Arshad Mansoor, president and CEO of EPRI. He added that these data "come from diverse sources, including regional system operators, power companies, research institutions, private equipment manufacturers, third-party analysts, and cloud and software vendors."

"It is estimated that 95% of data is locked behind power companies' cybersecurity and customer privacy barriers," added Swati Daji, executive vice president and chief financial, risk, and operations officer at EPRI. EPRI's Open Power AI initiative will "create, develop, and curate datasets as an open solution repository for the industry."

The initiative will also build "sandboxes for testing and validating use cases," as well as "feedback loops for sharing lessons learned," said Swati. Data in the open repository can be integrated with protected utility data for further AI modeling; as grid modernization advances, utilities will be better able to manage the rare peak loads that drive significant expenditures, she added.

Another initiative is the "One Digital Grid Platform," developed by component manufacturer Schneider Electric in collaboration with Microsoft and data provider Esri, serving as a "grid data hub," introduced Ruben Llanes, CEO of Digital Grid at Schneider Electric. The platform will "ingest data from utility smart devices" and "enable real-time insights, predictive analytics, and automation," he added.

Both initiatives will enhance utilities' ability to leverage AI-driven data from grid-edge intelligence to "disaggregate customer energy usage and coordinate flexible loads," said Menon. However, this requires "investment in distribution lines, software, and communication technologies," she added.

If planned holistically, "incremental investments in control room and grid-edge technologies can keep upgrade costs affordable," said Menon. An AI-enabled power system "might complete 10 million calculations in the time it currently takes to complete 10,000," supporting "more comprehensive planning and more granular, real-time system operations," she added.

Even with an expected 15 million customer-owned smart devices on Southern California Edison's system by 2045, AI tools can still monitor system data "and flag where critical decisions are needed for human operators," said Menon.

Policy efforts are accelerating to promote greater data access for all power system stakeholders.

Policy drives data access

Data accessibility has become a theme of recent federal, state, and utility policy initiatives.

At the federal level, a bipartisan AI framework proposed in 2023 by Connecticut Democratic Senator Richard Blumenthal and Missouri Republican Senator Josh Hawley, as well as the PREPARED for AI Act introduced last year by Michigan Democratic Senator Gary Peters and North Carolina Republican Senator Thom Tillis, both include data protection and sharing provisions, but neither has passed.

However, in 2024, utilities and regulators continued to address access to advanced metering infrastructure (AMI) customer data at the state level, according to the latest state policy action update from the North Carolina Clean Energy Technology Center.

According to the 2023 Grid Data Sharing Playbook issued by the National Association of Regulatory Utility Commissioners (NARUC), all power system data can be shared, but personally identifiable customer, demographic, or program participation data cannot. The playbook also notes that data access can yield system benefits, limit negative impacts, and clarify the distinction between "essential data" and "nice-to-have data."

Access to both grid data and customer data is problematic, said Chris Villarreal, founder and president of consulting firm Plugged-In Strategies and co-author of the playbook. Practices in the playbook can accelerate distributed energy resource (DER) interconnection and drive economy-wide electrification by speeding data access, providing transparency, and necessary customer data protections, he added.

Several states are undertaking significant data-sharing policy work.

According to a recent assessment by The Brattle Group for New York's "Future Grid" proceeding, New York's 2040 grid flexibility potential could avoid nearly $3 billion in annual power system costs. But realizing this flexibility potential "requires massive amounts of data, some of which is currently not measured or available," said Akhilesh Ramakrishnan, associate director of energy management at Brattle and co-author of the study.

Utilities must invest to "measure and collect" data and build "a single point of interaction and unified data formats" to simplify access, Ramakrishnan added.

In the first phase of the California Public Utilities Commission's (CPUC) R. 22-11-013 proceeding addressing DER growth, a data working group has been established to improve data access. The group will "develop rules and requirements for data access," including accuracy standards, customer privacy, and utility compliance with regulatory requirements, the group stated.

Accessible real-time data will enable all available resources to "be used to balance supply and demand without spending hundreds of billions of dollars on system upgrades," said Jon M. Williams, CEO of distributed battery provider Viridi.

In Massachusetts, lawmakers recently passed Senate Bill 2967, which requires "implementation of advanced metering data access protocols." DER advocates support the law, but utilities are seeking regulatory adjustments.

Eversource Energy "fully supports" data access and recommends a "common statewide portal using a virtual central data repository" rather than "legacy databases" that could "duplicate costs" and create "security vulnerabilities," said Jared Lawrence, senior vice president of customer operations and digital strategy and chief customer officer at Eversource.

In Minnesota, regulators approved "open data access standards" and approved Xcel Energy's $563.7 million investment to deploy smart meters and the wireless communications supporting them, said Annie Levenson-Falk, executive director of the Citizens Utility Board.

Utilities are also required to "collect and share aggregated or anonymized, disaggregated customer energy usage data," said Levenson-Falk. But Xcel rejected the initial data request, citing potential threats to customer privacy or security, and that rejection is currently being challenged, she added.

Other key state efforts are advancing in New Jersey, Arkansas, Illinois, and the District of Columbia.

A common focus of these proceedings is using AI-managed distribution system data to integrate aggregated DERs and virtual power plants (VPPs) into system operations.

Distribution system data is key

Stakeholders increasingly agree that investing in DER management systems (DERMS) will be necessary to access, share, and use distribution system data.

Each of the two types of DERMS is "a collection of multiple software systems that interact with utility operations and information technology systems and business processes," according to an April white paper from the Smart Electric Power Alliance (SEPA). Together, they can "enable a new paradigm of grid operations," and "no single software system can provide this capability," the white paper states.

Edge DERMS "manage a large number of behind-the-meter DERs from multiple DER manufacturers, such as batteries, thermostats, and electric vehicles," said SEPA. Grid DERMS "are tightly integrated with utility distribution operations systems, focusing on managing larger front-of-the-meter DERs as well as aggregated customer-owned DERs," it added.

"More actively managing and integrating DERs can improve grid reliability, power quality, and system resilience, and create economic value for utilities and customers," said SEPA. "Uncoordinated DERs can increase system peak demand, cause distribution overloads or voltage violations, and increase the risk of cascading outages," it added.

Grid DERMS costs may range from approximately $10 million to $20 million, while edge DERMS are typically under $10 million, depending on the complexity of the utility's system, said Lakin Garth, senior director of grid strategy at SEPA and co-author of the white paper.

If DERMS can provide compelling system data, utilities will be able to use DER aggregation and VPPs to provide the same flexible system services as other dispatchable resources—a view shared by utilities, DER advocates, and analysts.

Edge DERMS are already disaggregating behind-the-meter customer energy usage to address peak demand challenges and reduce distribution upgrade needs, said Jesse Demmel, chief technology officer at Uplight, a DER aggregator and edge DERMS provider.

"The data needed to provide more services is available, but it is siloed across utility planning, operations, and customer service departments, making it difficult to integrate into a single platform," Demmel added.

By investing in DERMS, emerging AI modeling can integrate "single pane of glass" platforms providing visibility connecting customers to utility control rooms, said Demmel. These investments will pay for themselves by using data to identify the most cost-effective upgrade options, he added.

"DERMS investment is absolutely necessary, along with investment in hardware, operations, and measurement technologies," and "DERs will benefit the system," agreed Brattle's Ramakrishnan. However, AI may not fully resolve utility concerns about safety or reliability, because attributing operator decisions to AI recommendations "would not satisfy regulators or customers," he added.

Plugged In Strategies' Villarreal agreed. AI "may enhance system modeling capabilities, but integrating all resources into a single one would be costly and computationally complex," he said.

However, AI operating at the distribution system level might eliminate data-sharing concerns because "it will follow the power flow, obtain the data needed to optimize system resources, and demonstrate why that data should be available," Villarreal added.