Angeli Mehta looks at how companies such as Verv, DeepMind, Vigilent, Novacab and Nnergix are employing machine learning to help them rise to the climate challenge
Climate change is our starkest challenge: could artificial intelligence help us meet it? There’s seemingly no aspect of the efforts we need to make that couldn’t be accelerated by artificial intelligence. In April, a report by PwC and Microsoft suggested that across four key sectors – agriculture, energy, transport and water – AI could enable a cut in global greenhouse gas emissions of between 1.5% and 4% by 2030, with its impact greatest in transport (up to 1.7%) and in energy (up to 2.2%).
AI could also potentially create between 18.4m and 38.2m net jobs across the sectors the report’s authors examined. However, these positive impacts depend on other innovations, such as distributed generation and storage, and an industrial internet of things (IoT).
AI itself requires large amounts of computing power, also requiring energy – and adding to the challenge.
Today’s AI systems are good at rapidly making sense of vast amounts of data – discerning patterns where we can’t
In a recent presentation, Mustafa Suleyman co-founder of Google’s DeepMind said: “Many of our most challenging problems are intractably complex. We’ve got tonnes and tonnes of data, but trying to extract insight from that data and learn the relationship between cause and effect well enough to make meaningful predictions ... is becoming more and more challenging.”
Today’s AI systems aren’t anywhere close to re-creating human intelligence, but they are good at rapidly making sense of vast amounts of data – discerning patterns where we can’t. Moreover, machine-learning algorithms can acquire knowledge from the data they analyse, so models become more accurate over time, helping humans to make better decisions. That might be to select the most fuel-efficient route for a ship (see Appliance of science), or to predict how weather systems will impact the output of a wind farm.
Suleyman wants AI to do good; to have an impact. The DeepMind team decided to pitch their efforts at two challenges: cutting energy consumption, and getting more renewable energy into the grid. First up was to use AI to extend the life of an android phone battery. A lot of battery power is wasted keeping apps up to date in the background: predicting which apps you’re likely to use soon, compared with those you might not look at for hours or even days, saves energy.
What if we could use AI to cut the energy we consume in using all our consumer or business devices? Smart meters are meant to help us do that, but can only show the total amount of electricity being consumed (at approximately 10 second intervals), so it’s not easy to work out which are the most energy-consuming devices.
UK firm Verv is changing that by applying machine learning to deduce which appliances are on in a household, and what they’re costing. Each electrical device has its own voice – an electronic signature that can be separated out through pattern-recognition technology. Verv’s technology can sample data one million times every second. “I liken it to having a microphone in a room and lots of people talking at the same time,” says Maria Kavanagh, Verv’s chief innovation officer.
She reckons Verv’s current system has the potential to cut energy use in the home by about 10%.
The use of energy will skyrocket with the advent of 5G applications like autonomous vehicles
Our connectedness is using vast amounts of energy. In 2016, Netflix’s indirect energy use (that’s you and me watching programmes downloaded from data centres) was 100,000 megawatt hours (MWh); by 2018 it had almost doubled to 194,000 MWh. This use of energy will skyrocket with the advent of 5G applications like autonomous vehicles.
In a recent blog post, Microsoft’s president Brad Smith said the company would launch “a new data-driven circular cloud initiative using the Internet of Things, blockchain and artificial intelligence to monitor performance and streamline our reuse, resale and recycling of data centre assets, including servers.” Microsoft’s AI platform Azure is already offering scientists new tools for monitoring the environment, and climate change mitigation.
Google has made much of its progress in using AI to cut data centre energy consumption, 40% of which goes on cooling. A three-year programme of analysis and learning by DeepMind produced a system that cut by up to 30% the energy used to cool its data centres. Over the course of the first year of deployment the system got better, and learned to take advantage of, for example, cooler winter weather; and to provide levels of certainty that taking specific actions would produce the desired outcome.
Vigilent, based in Oakland, California, is one of the signatories to the Step Up Declaration, an initiative launched by former UNFCC head Christiana Figueres at last year’s Global Climate Action Summit in California. The coalition of 22 tech companies, including Salesforce, Autodesk, BT, Cisco, HP, and Uber, pledged to harness technology to help reduce emissions across all economic sectors, starting with their own.
Vigilent has committed that, together with its data centre partners, it will “use AI to eliminate wasteful cooling in data centres and telecom facilities, cutting annual carbon emissions by 50 million metric tons.” By deploying AI globally, it anticipates carbon emissions could be cut by 10 times as much.
Cliff Federspiel, Vigilent’s president and chief technology officer, says Vigilent is in the process of extending its technology to commercial buildings, where efficiency measures are urgently needed. In the longer term, he says, “I do think the technology can be delivered to other types of process industries – food processing, pharmaceuticals, indoor agriculture by using the same algorithm.” But for now Federspiel believes data centres are where the biggest impact on carbon emissions can be made.
This enhances what people do – they don’t have to deal with minute-to-minute, hour-by-hour decisions
As for the fear that AI and machine learning will replace people, Federspiel thinks the opposite will be the case: “This enhances what people do – they don’t have to deal with minute-to-minute, hour-by-hour decisions on cooling. But they get data on where they can make more important decisions: for example, where to put IT equipment. AI eliminates those headaches and allows them to make better decisions.”
AI could also be important in helping the UK deliver on its ambitious target, announced last month, to be net-zero carbon by 2050.
The National Grid had already set a target for Britain’s electricity system to be zero-carbon by 2025. Meeting it will mean maximising the renewable energy fed into the grid from a vast array of producers – from individuals to industrial-scale producers.
One of the challenges is in anticipating how much renewable energy will be available – a problem by no means confined to the UK. Barcelona-based Nnergix is using AI to give grid operators, energy traders and producers around the world highly accurate energy forecasts for solar, wind and hydro production in the hours and days ahead. Its systems can also help investigate sites for their energy generating potential.
“Our data analytics are helping to avoid carbon emissions because the main goal is to enable the grid integration of renewable energy sources,” explains Joan Miquel Anglès, the company’s commercial director and co-founder. More advanced deep learning and better weather forecasting will only improve predictions. However, he adds, with climate change comes more volatile weather patterns: a further challenge for machine learning.
DeepMind, too, has been working on making wind energy more competitive with fossil fuels.
Intelligent energy storage is also going to be critical for building the electricity grids of the future
Last year, Google started applying machine-learning algorithms to part of its fleet of renewable energy projects. Using weather forecasts and historical turbine output data, it trained a neural network to predict what wind output would be 36 hours in advance.
This proved to be a complex task because output was so variable. The team is still refining the algorithm, but says the machine learning has boosted the value of Google’s wind energy by 20%, because it can now tell the grid in advance when and how much energy a given wind farm will deliver.
Intelligent energy storage is also going to be critical for building the electricity grids of the future, says Stéphane Bilodeau, chairman of Canadian firm Novacab.
The technology firm has developed hybrid energy storage systems, which generate electricity from heat and store it for when its needed.
So in buildings, energy can be stored during off-peak hours, and discharged during peak usage times to lessen the burden on the existing power grid. In a big power plant, instead of energy being lost through the cooling towers it could be stored. In transportation, fleets of delivery vehicles – which might themselves have to refrigerate goods – can maintain electrical power for heating and cooling without having to run the engine or a generator, so reducing fuel consumption and carbon dioxide emissions.
All these applications require analysis and prediction, precisely where machine learning comes into its own. Bilodeau reports that Novacab’s systems cut energy consumption by 14%-40%.
The technology is not that difficult ... smart products exist, and there are so many grants out there for innovation in this space, but it’s about getting policy change
A new manufacturing plant Novacab is building in upstate New York will get 80% of its energy from Novacab’s own hybrid systems. Bilodeau expects it may not have to go to the grid for power at all.
Back in the UK, Verv believes AI could be deployed to better predict electricity demand in order to keep the grid in balance.
“Think of each electricity substation … and imagine you have a battery there and a community could store its excess energy, and was incentivised to do so,” suggests Kavanagh. Add in smart-plugs, and AI could be used to enable the grid to manage sudden surges in energy demand by briefly switching off – say – all the fridges in an area rather than having to call on expensive back-up generation from fossil fuel plants. Where, when and for how long to switch off are the kind of rapid decisions that can be made using AI.
“The technology is not that difficult ... smart products exist, and there are so many grants out there for innovation in this space,” Kavanagh adds, “but it’s about getting policy change.”
She points out that consumers will need to have an incentive to share their data.
But the changing shape of the grid, made possible by AI, may ultimately lead to a fairer energy system.
DeepMind’s Suleyman sees enormous potential to radically improve current systems by working with existing data, hardware, and infrastructure
Verv has been supported by Ofgem to develop peer-to-peer electricity transactions. It made the first, using blockchain, in a pilot in London last year. The experiment centred on a community solar project that is powering the communal areas of blocks of flats in Hackney. Unused energy is exported to the grid at a rate of 5p per kilowatt hour.
Because Verv was monitoring energy consumption, it realised almost 80% of the energy produced was being sent to the grid, rather than the standard assumption of 50%, meaning that community, many of whom were paying 15p/KWh for their electricity through pre-payment meters, was being short-changed.
Phase two of the pilot will allow residents to be fully reimbursed for the energy they sell to energy firm Centrica. Refining Verv’s system could take six to 12 months, so it has full seasonal data to enable optimum decision making on when to store energy by charging up batteries, when to trade, or send energy to the grid.
Kavanagh expects this green energy-sharing platform could deliver at least a 30% reduction in carbon emissions. All this could be done with consumers completely unaware of the complex sets of decisions being made behind the plug.
DeepMind’s Suleyman sees enormous potential to radically improve current systems by working with existing data, hardware, and infrastructure.
Collecting the right data could bring about a step-change, he says: “In the next decade we expect really remarkable breakthroughs to come.”
Angeli Mehta is a former BBC current affairs producer, with a research PhD. She now writes about science, and has a particular interest in the environment and sustainability. @AngeliMehta.
This article is part of the in-depth Tech for Good briefing. See also: