Whether through impressively smart chat bots, automated driving or copywriting, artificial intelligence is a potential game changer in many industries. The electricity sector could also realize AI-powered efficiency gains, but the road to widespread implementation will be long.

AI-enabled technology could be used to make load and supply predictions, conduct inspections and maintenance, and manage load flow — all based on data analysis. In the case of an electricity grid, such data includes generation, demand and weather information.

"Smart grids are about how you can utilize existing grids and minimize the need for new investments," said Gerd Kjølle, director of the Centre for Intelligent Electricity Distribution, a Norwegian think tank. "Through the digitalization of grid sensors, you move from manual operations to self-healing grids, utilizing a diversity of technologies."

While the need for a smarter electricity grid is now widely accepted, the use of AI in crucial grid functions is still a long way off, according to Antonia Heinemann, senior manager at think tank Umlaut, part of the consultancy Accenture.

For one, the data required as input is subject to technical challenges. "In the best case scenario, this data is available in a clean and interoperable format," Umlaut analysts said in a 2021 report. "In reality, however, this is rarely the case, and the necessary quantity of data first have to be laboriously gathered from data silos in order for an AI system to be developed."

One source of data collection comes from the rollout of smart meters, which track household usage of electricity. But uptake rates vary significantly between countries.

Experts also see a shortfall in spending on sensors across the power grid. Grid remuneration generally encourages investments in physical infrastructure rather than software and smart tech.

"The bigger the asset base, the greater the allowed revenues and the opportunity to make returns," said Alex Schoch, head of flexibility at Octopus Energy Group Ltd. "This approach creates a very strong bias in networks to solve issues through investment in physical assets at the expense of using technology and smart solutions."

AI still embryonic

Deployment of new technologies also depends on grid operators' strategic outlooks, according to Kjølle. "They really need to see the benefit, and there is a cultural problem. Grid companies are monopolies, they are not competing in a market."

Initial steps are underway at Belgian grid operator Elia Group SA/NV, for instance, which is trialing AI to help detect signs of system imbalance that can inform engineers' decision-making. But experts see more promise in distribution.

In that space, Octopus Energy is already testing the potential of smart technologies at speed. Via its Kraken software, the company says its Intelligent Octopus plan leans on advanced data and machine learning — a branch of AI that uses algorithms to imitate the way humans learn — to automate a majority of the energy supply chain for customers. In return, customers get cheaper power by automatically shifting usage away from peak hours.

Energy storage developer Fluence Energy Inc., meanwhile, has AI-powered tools designed to improve the technical and commercial performance of renewables assets, which also includes an offering for bidding generation or storage into electricity markets.

These tools still require a lot of human knowledge and intervention, and as such are "still in embryo state," according to Gianmarco Pizza, global head of digital asset management at Fluence.

"Right now, we're using AI to make computationally tough decisions [and] to increase the speed at which decisions are made," added Gary Cate, a senior manager in electricity market design at Fluence.

Using such software could increase asset performance by up to 10% in a year, Pizza said. In the future, such tools could also become smart enough to make operational decisions, such as automatically scheduling maintenance visits at power plants.

Grids are no place for experiments

The success of self-contained AI ecosystems does not mean the technology is fit for adoption by the entire grid, according to Umlaut's Heinemann.

Such localized solutions "do not interfere with critical infrastructure, so you can deploy algorithms that are not 100% perfect," Heinemann said. "But when I think about grid operators, that is the heart of the system."

Heinemann sees a ramp-up of such systems in parallel trial modes over several years before widespread implementation in critical infrastructure. This can be done with grid simulators, for instance.

"[Grid operators] cannot be the ones running experiments," Pizza said.

According to Loïc Tilman, head of innovation at Elia, data is one limiting factor for automation. "We don't have enough data to train the model," Tilman said about the company's AI drive, which already includes some forecasting and automation functions. To that end, grid operators should commit to data sharing to improve quality, the head of innovation said.

In the longer term, Tilman expects many more grid functions to be automated and drew comparisons to a plane's autopilot setting. "You will have an autopilot for your grid which will not just be cool to have, but also mandatory."

Systems like ChatGPT, an AI-powered writing tool, are more immediately useful for deploying AI in energy companies, for instance in customer service, Heinemann said, adding that text and language recognition tools can also support tasks like processing permits for electric vehicle grid connections.

Utilities should consider their own vision for a future with AI, and their readiness for it, by investing in data management processes and also skilled engineers with a grasp of both electricity technologies and digital tech, Heinemann added.

The machine-learning community itself also still has much to learn before it can demonstrate its ability to tackle larger problems in the energy system and earn the trust of policymakers, according to Puneeth Kalavase, vice president of data science and engineering at Fluence. Energy "is a highly regulated industry for a very good reason, and [AI systems] are, for better or worse, black boxes," Kalavase said.

With or without AI, decentralization and digitalization of generation and power consumption raise the specter of cybersecurity risk, but experts see an in-built diversification of risk too.

"By increasing the amount of distributed energy assets and green generators on the system, grids are fundamentally deploying more resilience and building a grid that will be able to withstand more geopolitical and environmental pressures," Octopus Energy's Schoch said.

S&P Global Commodity Insights produces content for distribution on S&P Capital IQ Pro.