Utilities tend to approach new technologies with caution and skepticism compared with other industries, and artificial intelligence is no exception.
Despite that hesitancy, research from Capgemini Research Institute suggests energy and utility companies are already getting their feet wet with the technology: 39% say they’ve established a dedicated team and budget aimed at integrating generative AI, and 33% report they’re already working on pilot projects.
And, according to analysis from Indigo Advisory Group and Latitude Intelligence, generative AI is just the start — there are already more than 50 possible use cases across utility operations and DER integration for machine learning, computer vision, robotics, and predictive analytics
Avangrid, which owns and operates eight electric and natural gas utilities in New York and New England, is further ahead than most. In August, the company unveiled an in-house AI development team.
The seven-member team — made up of data scientists, engineers, and analysts — is developing proprietary AI systems for Avangrid, rather than focusing solely on integration. They will initially concentrate on developing three custom tools: A computer vision system to analyze equipment health (by identifying sagging wires or broken cross arms, for example); a platform that will leverage weather and grid asset data to analyze outage risk; and a predictive analysis model to determine health and life expectancy of equipment.
Latitude Intelligence analyst Fei Wang said the energy sector’s apparent interest in putting AI to work for the grid demonstrates a proactive embrace of the “new reality” of rapidly advancing grid edge technologies and the proliferation of distributed energy resources.
An investment in its own AI talent may help Avangrid adapt more quickly, she added, by allowing it to circumvent some of the challenges utilities face when looking to upgrade their technology stack.
“Most vendors will say their experience is that utilities tend not to really care about what the technology is, or about the nitty gritty of the technology stack,” Wang said. “As long as you get the job done, they’re very happy.”
Vendor solutions are typically siloed within a utility, she added; it’s rare that they have a foundational impact. Whether Avangrid’s in-house team can induce broader change will depend on how well its work is integrated across the wider company, and how widespread executive buy-in is.
Shifting priorities call for new strategies
Edward Mulholland, Avangrid senior director of performance and control center processes, said the decision to bring AI model development in-house is part of a broad effort to make data-driven decision making a “core competency.”
That’s not to say that Avangrid has completely cut ties with vendors — those relationships still exist for projects Avangrid can’t execute on its own, or to supplement internal work.
“But we really feel that we want to have this core in-house team that can drive the project, we can allocate resources to the needs that we have, we can react,” Mulholland added.
One reason to move more quickly is the changing regulatory landscape. “Regulators are wanting more detail on how the network is performing, how we are prioritizing investment, and so being able to be agile and respond to those requests and develop new models fairly quickly is one of the big benefits of having that internal team,” he said.
However, the road to implementing and scaling new technologies inside utilities is often a challenging one. There’s a split in the AI conversation between teams responsible for keeping the lights on and those tasked with innovation, explained Duquesne Light Company’s director of advanced grid systems and grid modernization Elizabeth Cook.
“It’s a tension-filled discussion because there is an urgency to keep the power on,” she said. “It’s really hard to defocus those who are in charge of doing that, and those who will need to drive this whole new way of thinking.”
For Avangrid’s AI team (and for the use of AI elsewhere in the energy sector) there’s also the hurdle of data availability.
While much of the data Avangrid is using to train and develop its proprietary models already exists — photos of poles and wires, information about outages, weather data — there’s not a lot of useful historical data.
There isn’t necessarily an out-of-the-box solution today.
According to Mulholland, that’s particularly challenging for Avangrid’s “GeoMesh” project, which will identify strengths and weaknesses of its networks in order to forecast grid performance under certain conditions. There isn’t historical data about unprecedented, climate change-induced weather events on which to train models, because those types of events are, well, unprecedented.
Take wildfire risk, for instance. While wildfires have yet to be a big concern for the Northeastern states where Avangrid is based, Mulholland said the company is “trying to keep ahead of the game and understand what risk exists.”
With limited data on how its assets could be impacted by a hypothetical New York wildfire, the team turned to data from other utilities around the country, combining information about historic outages in other regions with weather data for the Northeast as well as the company’s own asset data.
It’s a matter of using the technology to get out ahead of a potential crisis, Mulholland said, and figuring out how to mitigate negative impacts far in advance. While Avangrid can leverage existing analytics, he added, “there isn’t necessarily an out-of-the-box solution today.”
Navigating the talent challenge
You’d be hard-pressed to track down AI experts who also have experience in the energy sector, which means Avangrid had two options when it came to hiring for their new team — people with utility experience who would be trained internally on data science, or data science experts who would learn about the world of utilities on the job.
Avangrid went with the latter, hiring team members who hail from as far and wide as healthcare, astrophysics, and finance. None of them had ever worked in a utility.
There’s a certain level of risk in that hiring choice, and it was a “step into the unknown,” Mulholland added. But it’s a decision Avangrid is doubling down on as it looks beyond the team’s first seven members, and toward building a pipeline of talent that has both data science and energy sector experience.
To that end, Mulholland said Avangrid is leveraging its early career development programs, focusing on recruiting recent data science graduates who will help to grow the team from the ground up, gaining experience that can prepare them for the new types of roles required in the energy transition.
Despite the relatively small size of the team, Avangrid’s step into proprietary AI tools may push other utilities to begin thinking about how to future-proof their IT operations, said Wang.
“I think this at least shows that there is initiative to say ‘We’re not just adding applications on the side — we really want to impact how it works, how it’s structured,’” she added. But no matter how good the intentions, implementation will be a key element in whether this in-house model spreads.
“People love case studies, because they always want to know what other people did,” Wang said. “I think it’s always good to see a utility lead these efforts, because of the conservative nature of the industry.”
Mark Waclawiak, senior manager for operational performance at Avangrid, will be presenting at Latitude Media’s Transition-AI: New York conference on October 19th. See the agenda here.
