Around 50% of the shift from fossil fuels to renewable form of energy has been a digitised transition to 2020, says Energinet, Denmark’s power system operator. As the world digitises further, the already vast volumes of data will only increase in mass. Advances in Artificial Intelligence technology and machine learning tools provide ways of processing this tsunami of data into a new world order
The energy industry is starting to understand where artificial intelligence can be applied effectively
GRID CONTROL Artificial Intelligence could be a requirement as transmission network operation becomes increasingly complex
DATA MASS Storage and management of the rapidly increasing levels of data will add 20% to global energy consumption
KEY QUOTE One day AI will be everywhere because it will be impossible to manage assets in a sophisticated grid without really addressing the relationship between infrastructure and environment
Given the accelerating electrification of the world and aggressive carbon emission reduction goals, Artificial Intelligence (AI) and machines that learn by doing will help develop next-generation power systems running mainly on variable renewable energy, expand the use of electric vehicles (EVs), increase plant-scale storage capacity, and do more with as few new transmission lines as possible.
AI and machine learning are terms that are—inaccurately—used interchangeably. Simply, AI can track vast quantities of data and is taught to detect patterns not easily discernible to humans. Using AI increases the speed and precision of data-hungry programming. Computers make fewer mistakes and do not get tired. Today, AI can often analyse an X-ray better than a real-life technician.
Machine learning, meanwhile, is a subset of AI allowing devices to learn from data without being programmed explicitly. Put another way, machine learning is the optimisation of data. For the grid, AI can help with electricity sector asset management, control and acquisition, according to Shuli Goodman, executive director of California-based Linux Foundation (LF) Energy, an energy software non-profit organisation.
Kristian Ruby, secretary general of Eurelectric, a lobbying group for European utilities, says society will eventually move “beyond digital” to form an ecosystem of interconnected technologies. For next-generation power systems, decentralised machines and machine learning must be adopted, adds Ruby, supported by vast banks of memory processing using cloud storage and new 5G technology.
But it is still early days for AI’s involvement in managing the grid and generation assets. No utility has enough data, yet, to track how a particular vendor’s transformer is performing, when it needs maintenance or is heading for trouble. Tens of thousands of transformers would need to be tracked with AI to get an accurate assessment.
Luke Witmer, of Finland-based Wärtsilä, a power plant and storage provider, says AI is good for forecasting and decision-making, detecting anomalies in millions of data points. AI will be especially good at forecasting renewables’ variability at the grid and plant level, easing a faster roll-out of clean energy, and for fine-tuning predictive maintenance of power plants, he adds.
From a transmission operator’s point of view, AI will be vital as the next-generation grid becomes more than ever a cyber-physical system of systems. The complexity to control such an approach is high. RTE, a French transmission system operator (TSO), says it needs to coordinate a large population of agents and devices with some partial autonomy.
“Digitalisation and AI seem promising and perhaps mandatory to tackle this task,” says the operator. “They represent great opportunities especially when trying to use different ways to solve grid congestion. For example, the adaptation in real-time of the grid topology or the connectivity between components in the grid,” it adds.
The grid operator has already been using AI, especially machine learning, in different industrial processes, such as load and wind or solar generation forecasts. “Many decisions are assisted by tools using intensive simulations and some expert rules analysis, a sort of AI,” the operator adds. Yet there is a single major challenge in the use of AI—it lacks common sense and intuition and cannot certify whether its answers will always be accurate. A second challenge with AI is that there is typically a lack of transparency in explaining the algorithms used.
BRAINS AND BRAWN
Eurelectric’s Ruby believes that grids are already starting to be characterised by far more complex flows of electrons to accommodate decentralised generation and storage, in contrast to the one-way flow of electricity of the past. “Electric cars and heat pumps will talk much more to the grid,” accelerating the need for more AI, he notes. The next-generation neural grid—equipping the grid with an intelligent “brain”—will need not only to be stronger but also cleverer at every point of distribution and consumption, says Verizon Wireless, a US-based telecom company.
It looks likely that 5G—the fifth generation of cellular networks—and AI could be employed in supporting our daily needs such as controlling heating in buildings automatically based on usage or determining when an EV can charge most cheaply given the owner’s timetable. It will also be crucial for managing consumer profiles and the peaks and troughs of demand, such as when consumers get home from work and want to charge their EVs.
Using EVs to feedback into the grid would require significant infrastructure changes, including reverse-flow metering and calculating compensation to a car owner or renter for when their vehicle supplements a utility’s power generation. Using AI could help keep track of and control battery stores without the need for physical upgrades.
Beyond managing grids, generators and consumer demands, AI can also be used to avert or minimise a transmission network disaster. On January 8, 2021, an AI-aided mechanism based on contracts with large industrial consumers—which receive a signal in the event of a large frequency drop to reduce their consumption—helped save the European electrical system from having to shed load after a problem in Croatia that had split the continental European grid in two areas, according to RTE.
Similarly, in the US, after Hurricane Irma tore through South Florida in 2017 and cut power to more than six million people, it took ten days for the local utility Florida Power & Light (FPL) to switch the lights back on. FPL is now investigating how AI could be used to reduce recovery time if the situation arises again. FPL wants to use sensors and drones to locate outages and decide how to better pinpoint problems and resolutions.
According to RTE, using AI will require new definitions of reliability and resilience of large power systems, leading to more advanced principles for risk management. AI could certainly help in this new context, RTE says.
The value of AI is that huge amounts of data that relate to the grid are already being collected. This trend will accelerate as more products are digitised, sensors become smarter and electrification increases.
“We have a massive tsunami of data,” says LF Energy’s Goodman, storage and management of which will add 20% to global energy consumption, she adds. Utilities in a previous generation completely controlled power generation and they remain powerful and fearful of change.
Even the smart grid—which was first introduced 25 years ago—was sometimes spoken of as something in the distant future by some of those working at utilities just a few years ago. “We have a logjam of data, yet we have to move at the speed of technology. We need to throw dynamite into the logjam,” says Goodman.
The vast quantities of data in the electricity sector remain siloed but AI will help shatter those barriers to better inform grid managers, power plant operators and consumers, she continues.
French TSO RTE agrees: “AI could also help to detect inconsistencies [while] merging the data from different silos. The potential of AI can give more incentives for transmission system operators to break these silos, in particular between asset management and grid operation.”
Some AI-supported grid projects are already underway. National Grid, the UK group that also operates in the north-east United States, is using intelligent digital substations and electrical devices to ease integration of clean energy, provide better visibility of the system and enable predictive maintenance, says John Lamontagne from the company.
Over the next ten years, National Grid hopes to install more than 150 AI-powered digital substations in the US. The advantages of the substations include smarter investment and maintenance decisions, as well as data-driven decision making. The final phase of a pilot project in New Hartford, New York, was to be completed in the first half of 2021.
National Grid leveraged industry-specific standards to automate the substation’s electric power equipment while maintaining the safety, security and reliability required for the power grid. The digital substations include online monitoring capabilities to gather continuous real-time data on assets.
This real-time information will enable National Grid to make smarter, data-driven investments and maintenance decisions, it says. Using this approach, the company will be able to provide better value for customers by increasing system reliability, allowing it to focus instead on investments and maintenance projects.
“With this [pilot] substation and future ones, we expect to gain efficiencies and deliver cost savings for our customers and ensure even greater system reliability and resilience, all while more easily integrating additional renewable energy sources onto the power grid,” says Rudy Wynter, president of National Grid in the US.
GE, a US engineering conglomerate, uses a Digital Twin product, an analytic engine for optimizing operations at power plants that is boosted with AI. The models can show design limits of a power generation unit when commissioned or they infer the design limit for an existing single plant or array of units by matching the equipment to thousands of other similar pieces of equipment in the database.
The model, which improves as it learns, can accurately represent the plant or fleet under many operational variations—fuel mix, ambient temperature, air quality, moisture, load, weather forecast models and market pricing. The information is especially useful when managing the variable energy output of wind and solar units.
LineVision of Boston, which secured financing from National Grid in April 2021, has developed a non-contact overhead power line sensor that uses AI and cloud-based analytics and lasers to monitor, optimise, and protect energy delivery infrastructure. It gives dynamic line ratings for transmission lines, a real-time forecast of their power carrying capacity.
This helps utilities to optimise their grid. Grid congestion is an issue that must be tackled if President Joe Biden’s goal of a net-zero grid—with massive deployment of renewables—is to be achieved, notes Hudson Gilmer, CEO at LineVision, pointing to the grid congestion caused by the rapid roll-out of renewables plants.
Gilmer refers to the results of a study by the Brattle Group released in February 2021 for the US-based Watt Coalition, a collective of companies in the energy transmission sector working to add new grid technologies of which LiveVision is a member.
It found that three AI-backed technologies—dynamic line rating, advanced power flow control and topology optimisation—could enable Kansas and Oklahoma to integrate 5.5 gigawatts (GW) of wind and solar generation currently in interconnection queues by 2025, more than double what is possible without the technologies—and which would otherwise need to wait for more and expensive transmission capacity to be built.
An over-reliance on AI, however, can lead to problems down the line. A piece of incorrect programming could take ages to be discovered. Data can be ignored, without human intuition and common sense to interpret correct results. “This concern is at the heart of RTE’s research on the use of AI. We see AI as a promising [tool] to design an assistant, a tool that would bring the right information to the right person at the right time to make the best decision,” RTE says.
Furthermore, AI will be used with other decision-making tools and not as an autonomous agent. “One of our important research questions is how to mix classical methods from mathematical programming based on physics and machine learning,” it says.
Quantum computing, which would harness processing power to allow AI to utilise deep thinking is currently impractical. It has been looked at for decades and still remains years from being commercialised. In the short term, since energy utilities are mired in institutional inertia, they will surely keep human overseers in the loop long after the AI pilot projects have proven their worth.
“Even with AI, there’s no set it and forget it,” agrees LF Energy’s Goodman. “We have to be looking, we have to be watching. It has to be data scientists’ jobs to look at the data and understand it and be able to visualise what is happening.”
Similarly, AI can help ward off cyberattacks by quickly recognising unusual activity, say experts. The digitised grid is increasingly vulnerable to attacks, given that remote wind and solar farms are often unmanned.
In the case of a cyberattack where AI is deployed, RTE says the technology will not act directly on the grid and that many checks—such as a power grid simulator—will be performed and will propose responses. “We believe that this will limit the impacts of possible cyberattacks targeting specifically the AI,” it said.
Broadly, AI will come into its own when the flow of electricity changes from the traditional one-way passage—from generator to consumer via the grid—to where it becomes more complex. The increase in local energy consumption from distributed resources and reduction of load on the main grid. “Solving these challenges will be a difficult task,” says RTE. “We don’t plan on relying solely on AI to tackle these challenges.”
Rather, what RTE hopes is to use AI to help the operators in their daily routine, so they can focus more on these new issues that will require lots of expertise and adaptation to new situations. A partial solution will be to rely on automatic decentralised controllers with specific tasks.
“The operators will mostly become navigators, defining optimal trajectories for these controllers and will also be able to take control in case of extreme conditions. An assistant using AI will help them in this task,” the operator says.
Goodman also considers embedding ethics into AI regarding access to energy. “If we do not, it could have really severe social effects. It is such a holistic system,” she says. Connected devices in homes are an expense some may not be able to afford. More affluent people may have a Tesla in the garage and two Powerwall AC battery systems and solar rooftop, but the have-nots cannot afford it, meaning they risk being cut off by grid operators.
“There’s a very real possibility that we are going to create perverse effects that are going to penalise poor people and communities that don’t have resources [for AI] and [thus] create energy deserts,” Goodman says, referring to the gaps on the energy system if an operator becomes too reliant on domestic AI controls.
AI remains in its infancy, only commercialised for the grid in small and specific applications. But, Goodman says these projects are harbingers. “One day AI will be everywhere because it will be impossible to manage assets in a sophisticated grid without really addressing the relationship between infrastructure and environment. We are at the bottom of a ten-to-15 year journey with data, AI, automation and virtualisation, and there’s really no way to understand what it’s going to be like when we get to 2040.”
“We’re just on the verge of take-off,” adds Eurelectic’s Ruby. •