15 minute read 24 Nov 2020
Aerial view of solar power station

Why artificial intelligence is a game-changer for renewable energy

By Thierry Mortier

EY Global Digital & Innovation Lead for Energy

Innovative and creative leader. Curious accelerator and visionary. Technology enthusiast interested in emerging technologies, eMobility and green tech.

15 minute read 24 Nov 2020

RECAI 56: The low-carbon transition will need AI to integrate a large increase in intermittent renewable energy while ensuring a stable grid.

This article is part of the 56th edition of the Renewable Energy Country Attractiveness Index (RECAI).

In brief
  • Artificial intelligence (AI) has the ability to unlock the vast potential of renewables. Failure to embrace it means risking falling behind the curve.
  • The powerful prediction capabilities of AI will lead to improved demand forecasting and asset management.
  • The automation capability of AI can drive operational excellence in many crucial areas.

The energy sector faces pressing challenges and needs to act with urgency. Policy commitments to a net-zero future, such as the Paris Agreement, mean the transformation to a low-carbon economy must come at pace.

Major disruption to the electricity sector is on the cards as governments ramp up renewables and transition away from fossil fuels. While renewable energy looks set to flourish amid this backdrop, its intermittent nature means solutions will need to be found to keep grids stable. Additionally, the industry is changing from a market based on commodity pricing to a market based on technology solutions in order to integrate renewable energy. As the energy industry continues to utilize more variable generation sources, accurate forecasts of power generation and net load are becoming essential to maintain system reliability, minimize carbon emissions and maximize renewable energy resources.

As we move into the Fourth Industrial Revolution, grid operators, developers and consumers are harnessing artificial intelligence (AI), paving a path for a smooth transition to a greater use of renewables. AI’s ability to provide better prediction capabilities is enabling improved demand forecasting and asset management, while its automation capability is driving operational excellence – leading, in turn, to competitive advantage and cost-savings for stakeholders.

Supported by other emerging technologies, such as the internet of things (IoT), sensors, big data and distributed ledger technology, AI has the ability to unlock the vast potential of renewables. Failure to embrace it would leave the renewable energy sector falling behind.

AI is far superior to humans when it comes to carrying out complex tasks at speed. Given that an energy grid is one of the most complex machines ever built and requires split-second decisions to be made in real time, AI algorithms are a perfect fit.

(Chapter breaker)
1

Chapter 1

How AI is transforming renewable energy

From demand forecasts to asset maintenance, the application of AI could bring gains on many fronts.

As an increasing amount of megawatts feeds into the grid from variable renewable energy sources, predicting capacity levels has become paramount to secure a stable and efficient grid. This is due to the fact that with renewables taking up a greater share of the grid, there is a loss of baseload generation from sources such as coal, which provide grid inertia via the presence of heavy rotating equipment such as steam and gas turbines. Without grid inertia, power networks will be unstable and susceptible to blackouts. Now, with the application of sensor technology, solar and wind generation can provide an enormous amount of real-time data, allowing AI to predict capacity levels.

Before harnessing AI, most forecasting techniques relied on individual weather models that offered a narrow view of the variables that affect the availability of renewable energy. Now, AI programs have been developed – such as IBM’s program for the US Department of Energy’s SunShot Initiative – which combine self-learning weather models, datasets of historical weather data, real-time measurement from local weather stations, sensor networks and cloud information derived from satellite imagery and sky cameras.

The result has been a 30% improvement in accuracy in solar forecasting, leading to gains on multiple fronts. “We found that improved solar forecasts decreased operational electricity generation costs, decreased start and shutdown costs of conventional generators, and reduced solar power curtailment,” says Hendrik Hamann, Distinguished Researcher and Chief Scientist for Geoinformatics at IBM.

Forecasts of the base variables – wind speed and global horizontal irradiance, as well as the resulting power output – allows for a view on a range of time horizons, from minutes and hours ahead (for maintaining grid stability and dispatching resources) to day-ahead (optimizing plant availability), to several days ahead (scheduling maintenance).

With increasingly larger data sets becoming available, predictions can now go far beyond the weather to train algorithms to predict more remarkable outcomes. For instance, how much additional power is used during a festive holiday, a large-scale international event, or how much altitude impacts a community’s energy use.

For generators and energy traders, more accurate forecasting of variable renewable energy at shorter timescales allows them to better forecast their output and to bid in the wholesale and balancing markets – and, importantly, to do so while avoiding penalties.

“The earlier and more accurately you can predict, the more efficient it is for energy traders to rebalance their position. I see AI providing a way of dealing with lots more sites and using more granular and diverse data than historic forecast methods,” says Alex Howard, Head of Strategy at Origami. “Ultimately, that means making a better financial return.”

For generators and energy traders, more accurate forecasting of variable renewable energy at shorter timescales allows them to better forecast their output and to bid in the wholesale and balancing markets – and, importantly, to do so while avoiding penalties.

Meanwhile, for grid operators, AI algorithms with vast amounts of weather data can ensure optimal use of power grids by adapting operations to the weather conditions at any time. More accurate short-term forecasting can result in better unit commitment and increased dispatch efficiency, thereby improving reliability and reducing operating reserves needed. 

“Now, with AI, we can predict more accurately what renewables are likely to do, so we can control other power plants more accurately, like coal plants that take many hours to ramp up,” says James Kelloway, Energy Intelligence Manager at National Grid ESO.

In turn, cost-savings can be passed along. He adds: “What we want to avoid is turning the renewables off. From a price-tag perspective and the way the system is configured, renewables are not only greener, they are usually cheaper.”

Through a grid-stability lens, with AI ensuring that the power grid operates at optimal load, grid operators can optimize the energy consumption of consumers. But it is not only transmission system operators that can utilize AI; its application goes beyond central planning and can play a bigger role on the edge of the grid with machine-to-machine communication. In an ideal situation, electricity generated within a neighborhood grid or solar PV system can be used to improve reliability and combat grid congestion – which is associated with complex, decentralized systems with bi-directional electricity flow.

The earlier and more accurately you can predict, the more efficient it is for energy traders to rebalance their position. … Ultimately, that means making a better financial return.
Alex Howard
Head of Strategy, Origami

Equally important is accurate demand forecasting – and here, too, AI has a key role to play given its ability to optimize economic load dispatch and improve demand-side management. Increasing installation of smart meters has enabled demand data to be sent to utility providers. AI algorithms can absorb the data, which can be sent as frequently as hourly, and predict network load and consumption habits accurately.

For consumers, utility bills can be reduced, with AI systems predicting a building’s thermal energy demand to produce heating and cooling at the correct times through optimization of home solar and battery systems. Efficiency gains are combined with load shifting to times when electricity is cheapest, with renewable electricity available in the system.

We can now predict when demand spikes will occur and discharge energy to keep customers’ grid-supplied electricity below a certain set point, and, consequently, help customers control energy costs without interrupting operations or requiring any involvement on their part,” says Josh Lehman, Senior Director of Product Management at US energy storage firm Stem. He adds that the company’s AI-driven software has improved customer savings by approximately 5% year-on-year.

In the all-important flexibility jigsaw, the ability to understand consumers’ habits and actions creates greater flexibility in a smart grid because AI algorithms can make predictions about a building’s energy use 24 hours in advance, based on its experiences in the past.

Battery storage also has an important role to play in providing demand flexibility, with AI again playing a pivotal part. As storage batteries can be activated quickly and used to manage excessive peaks – as well as minimize the back-up energy needed from diesel generators, coal-fired power plants or other gas-fired “peaker” plants that are utilized at peak demand – AI can predict and make energy storage management decisions by considering forecast demand, renewable energy generation, prices and network congestion, among other variables.

Battery owners can deploy their storage pack according to the compensation for the services provided by the battery. Stem has developed AI algorithms to map out energy usage and allow customers to track fluctuations in energy rate to use storage more efficiently.

In the all-important flexibility jigsaw, the ability to understand consumers’ habits and actions creates greater flexibility in a smart grid because AI algorithms can make predictions about a building’s energy use 24 hours in advance, based on its experiences in the past.

Similarly, US-based software-as-a-service platform provider AMS uses AI in versatile battery storage systems to optimize opportunities to purchase electricity from the grid when prices are low, and then sell back to the market when prices are high. Another case is Australia’s Hornsdale battery, with 150MW, which operates an autobidder AI algorithm, developed by Tesla, that has allowed the project to capture revenue streams about five times higher than an energy trader, according to AMS.

For electricity providers, AI can also assist with operations and maintenance of asset management. AI algorithms can automatically detect disturbances in real time of mechanical failure, thereby improving reliability and efficiency in the power system. By using data from sensors, algorithms can learn to distinguish and precisely categorize normal operating data from defined system malfunctions.

“Unexpected disruptions across the industry can cost 3%–8% of capacity and US$10b annual lost-production cost,” says Brian Case, Chief Digital Officer at GE Renewable Energy. Its Predix software is embedded with AI-based algorithms that can interpret industrial data to make predictions on machine health and recommend actions to improve efficiency for assets such as wind farms.

AI’s ability to root out system malfunctions immediately can also prevent a chain reaction. For instance, if one power plant should fail, an abrupt spike can be expected in the load placed on other power plants. This, in turn, slows down the generators, and the frequency decreases. If the frequency sinks below a threshold value, the operator may be required to cut off sections of the grid to maintain system stability. The ability of AI algorithms to make split-second decisions allows for appropriate, fully automated countermeasures to be taken.

Unexpected disruptions across the industry can cost 3%–8% of capacity and US$10b annual lost-production cost.
Brian Case
Chief Digital Officer at GE Renewable Energy

Finally, at a regulatory level, AI unlocks legislation to be created more effectively. It also provides insight into human motivations tied to renewable energy adoption and how consumer behaviors could possibly be changed to optimize the energy system.

(Chapter breaker)
2

Chapter 2

The challenges of applying AI across the sector

Poor data, consumer mistrust and regulatory barriers could all prove problematic for the technology.

AI’s potential to be a game-changer for the renewable energy sector is undeniable, but that does not mean its greater application across the sector is devoid of challenges.

In today’s digital age, concerns have emerged that relying on AI too much could leave energy networks vulnerable to cyber attacks. A wake-up call came in 2015, when hackers took 30 substations offline in Ukraine, leaving 230,000 people in the dark for six hours. A second, much smaller attack occurred on a transmission station a year later, in Kiev. It is believed the 2015 attack required months of planning and a team of dozens working in coordination and was largely due to the fact employees fell for a phishing campaign.

Another type of cyber attack on power grids that has been deployed more recently involves exploiting vulnerabilities in firewall firmware. In 2019, the North American Electric Reliability Corporation revealed that the first attack on a US grid network occurred with an undisclosed utility suffering communication outages between its control center and generation sites. The disruption was a result of an outside party rebooting the company’s firewalls. Each communication failure lasted less than five minutes, but the entire attack went on for about 10 hours.

However, the likelihood of another successful large-scale attack appears minimal. Operational technology (OT) systems are isolated from information technology (IT) systems, with no network connections between the two, and are, therefore, much more difficult to infiltrate. In addition, OT systems are more customized and esoteric, so they are far less familiar to would-be hackers.

If hackers did get into operations networks, they would need to learn the equipment and setups. Moreover, whatever equipment setup a utility may have, its physical processes can require real expertise to manipulate, as well as months’ more effort and resources. Ultimately, experts believe the vast majority of grid-penetration incidents will amount to little more than spear phishing.

From a performance perspective, data bias, audit and ongoing verification of algorithms are issues that AI systems must consider when developing algorithms. Machine learning is ultra-sensitive to poor data, with the adage “garbage in, garbage out” holding true here. It is critical that data is taken and made machine readable, so that it is quality in, quality out. For trusted AI, frequent verification of data is a necessity to ensure algorithms remain valid over time and that as the machines learn they do not deviate from the original algorithms.

Concerns have emerged that relying on AI too much could leave energy networks vulnerable to cyber attacks. However, operational technology systems are isolated from IT systems, with no network connections between the two, and are, therefore, much more difficult to infiltrate.

That is not always as easy as it sounds, however. “AI may have some limitations in areas that don’t have the historical data available to find the intelligence, because it has never occurred or existed before,” warns Hamann, at IBM. “You can, though, overcome these challenges by using different, more, and more selective data sources, as well as different techniques.”

From a technology perspective, reliance on cellular technologies would limit AI’s potential in rural and other under-served areas in many emerging markets, particularly low-income ones. Smart meters rely on constant data communication, so a lack of reliable connectivity is a substantial impediment in areas where cellular network coverage is sparse or limited.

As with all new technology, AI is likely to face initial mistrust from consumers. Building owners and occupants are likely to be skeptical that the technology can deliver reductions in either energy consumption or cost without compromising energy services and comfort. Robust education and marketing programs will be needed to convince customers to trust the technology.

Regulatory barriers also exist, including the fact that energy markets’ rules must permit the trading of flexible demand at a scale that allows commercial buildings to participate in the market. In some energy markets, for instance, the minimum allowable bids for participating are higher than the size of flexible loads likely to be offered by commercial buildings. In addition, some energy markets require access fees for participation, which might pose a barrier to entry for small-scale participants.

(Chapter breaker)
3

Chapter 3

Determining project viability with AI: a case study

How one company’s use of image recognition is providing crucial verified data for investors and developers.

For investors and developers, the million-dollar question is whether a prospective project will be financially viable. By using AI, that question is now easier and quicker to answer than ever before.

Investors and developers have long been hamstrung by the opacity surrounding energy data. There is no shortage of open-access data on grids and power-generation assets in the public domain, but developers need validated data to identify technical and financial risks for a prospective project to be profitable.

This is where image recognition – a rudimentary form of AI – can play a huge role in validating open-access data. ENIAN, a UK software firm, boasts one of the world’s largest renewable energy project databases, having gathered publicly available data on power plants and grid assets, and their coordinates. The company uses a matching script to run across the data sets, and the AI is trained to recognize what a wind turbine looks like. The AI takes millions of sets of wind-turbine coordinates, and identifies which images and coordinates match up.

AI can then be used in cost prediction, with an enterprise platform simplifying project managers’ workflow. With trustworthy data retrieved from image recognition, the platform provides datasets detailing assets’ grid connections, distance to nearest substation, existing power-generation assets in the area, and indicators on the performance of assets.

Investors and developers have long been hamstrung by the opacity surrounding energy data. There is no shortage of open-access data on grids and power-generation assets in the public domain, but developers need the ground truths to identify technical and financial risks for a prospective project to be profitable.

Projects are augmented by AI and managers can examine qualitative details such as how many competitors are in the area and if other projects in the area have failed – and, if so, why. The data also reveals solar irradiance and wind speeds for potential sites, as well as an optimal route-selection estimate for connecting a power-generating asset to the grid, so project managers can make quick and accurate models of what a solar or wind farm can yield. An algorithm then produces a preliminary cash-flow model that indicates whether a project is worth pursuing further.

These tools could open up frontier markets – such as Central Asia, for instance, where there is tremendous potential for wind and solar farms – but development has lagged behind. “Very little is known about the grid networks of these locations, and the data you may find is outdated and fragmented. This tool allows us to run a scan and see what is there, and work backward after identifying the site,” says Phillip Bruner, ENIAN CEO.

“There is so much opacity around where, actually, the nearest points of interconnection are, a project’s available capacity, and who the owner is, so having this data without having to rely on third parties is very useful.”

For developers, the name of the game is operational excellence, and automation through AI can allow firms to get a leg up on the competition by identifying profitable prospective projects quicker. The ability to rapidly scale data collection and analysis through automation also frees up time for project managers to focus on getting deals sealed faster, projects started earlier and timelines moved forward. 

By augmenting project managers with verified data, projects become more predictable, efficient and cost-effective, and, ultimately, lead to better returns on investment.

While AI has been criticized for its impact on the labor market, focus should be shifted to its ability to free up skilled labor from linear tasks. As Bruner puts it: “AI is coming for the tasks you hate to do, such as spending lots of time processing data and validating data. The machine can step in and do those mundane tasks, so it augments everyone’s capabilities.”

(Chapter breaker)
4

Chapter 4

Increasing research and development to bolster AI’s capabilities

Further R&D could find solutions to AI’s limitations and cut costs – just as it did with solar.

AI and its sidekicks of emerging technologies, including IoT, sensors, big data and distributed ledger technology, are game-changers for the renewable energy sector. Key accelerators such as prediction capability through demand forecasting and asset management, combined with increased automation providing operational excellence, are already leading to major cost-savings, better yields and improved returns on investment.

Just as R&D in the solar industry has driven down prices, further R&D in AI has the potential to lower costs drastically, while its capabilities should grow and solutions to its limitations could emerge. Governments are realizing this, too, with the US Department of Energy announcing in August 2020 US$37m of R&D funding for AI. The UK is also funding several new research hubs that will be created to develop robotic technology to improve safety in offshore wind.

“We are now at the point where the most sophisticated market participants are turning proofs of concept into real, scalable applications of the technology,” says Howard, from Origami, adding that he expects these applications to de-risk the area for others.

We are now at the point where the most sophisticated market participants are turning proofs of concept into real, scalable applications of the technology.
Alex Howard
Head of Strategy at Origami

“If we are going to reach a net-zero future, the grid needs to be a lot smarter,” says Bruner, from ENIAN. “It needs to be able to adapt to a lot of different power-generating and power-consuming devices that are interconnected, and that is where AI has the most potential to help renewable energy grow.”

For a net-zero future, AI could be the missing piece of the flexibility jigsaw. Its ability to ensure an efficient and stable grid will be paramount as an increasing amount of renewable energy floods into the grid.

Summary

AI is revolutionizing the renewables industry, with improved demand forecasting, superior asset management, and operational excellence through automation. As the low-carbon transition gains pace, AI’s ability to seamlessly integrate an enormous increase of intermittent renewable energy will be needed to ensure a stable and reliable grid.

About this article

By Thierry Mortier

EY Global Digital & Innovation Lead for Energy

Innovative and creative leader. Curious accelerator and visionary. Technology enthusiast interested in emerging technologies, eMobility and green tech.