When Ontario’s energy ministry launched its “green bank” this past summer to help homeowners and small businesses become more energy efficient, one of its first actions was to give away and install 100,000 smart thermostats.
On the surface, it seemed like an expensive publicity stunt. But dig a little deeper and the campaign, with its $40-million price tag, is arguably one of the biggest efforts in Canada to give artificial intelligence (AI) a foothold in the home.
Chances are, the thermostat hanging on your wall right now isn’t much of a thinker.
It turns on and off when you tell it to, but doesn’t really “know” anything about you. It doesn’t sense and react to its surrounding environment.
Smart thermostats, on the other hand, are always learning.
These clever devices, like models sold by Toronto-based startup ecobee or Google-owned Nest, never stop collecting data on local weather, and they’re constantly taking note of when we come and go from our homes. They track the times we go to bed and wake up and, with the help of occupancy sensors, know which rooms we tend to use most and when we typically use them.
The more data that flows into these wall-mounted gadgets, the more accurate they get. Using machine-learning algorithms to tune into our changing routines and behaviours, they use the least amount of energy possible to keep us comfortable in our homes.
It’s just a small taste of how AI is starting to transform our relationship with energy as we make the transition to a low-carbon economy — from how and when we consume electricity and fuels in buildings, vehicles and industry; to how we produce, deliver, store and even trade it.
The impact will be felt most in the electricity sector, which is expected to become more efficient, reliable, secure and safe as AI algorithms play more critical roles in an increasingly complex show.
“It may lead to a world where power generation, distribution and transmission operations are automatically optimized, where the grid is balanced independently of any human interventions, where trading and arbitrage decisions are made in nanoseconds at a scale that only machines could tackle, and where [customers] never have to worry about searching for a better supplier or changing the temperature manually,” McKinsey, the international consulting firm, envisioned in a discussion paper last June.
It’s unclear when the power grid will no longer need humans, but AI is already having a measurable impact.
General Electric says it can boost energy production from wind farms by as much as 20 per cent using machine learning to anticipate and better respond to wind direction and speed, and to monitor wear and tear on parts, allowing for proactive maintenance.
Google has slashed 15 per cent from its power bill by employing AI alongside a network of data-collecting sensors to make better use of fans, cooling systems and other equipment in its data centres.
In Hungary, a utility called Eon used machine learning to analyze the power consumption and bill-payment history of customers in a town where rising prices were encouraging people to steal electricity. Eon was able to spot likely suspects, and worked with a local charity to identify those in need of social assistance.
Own an electric car? If companies like Microsoft and GE have their way, AI could one day be your own personal power broker, watching the electricity market and charging your vehicle only when the price is low or the wind is blowing; then selling the electricity in your car battery back to the grid when it can fetch a higher price (and you don’t need it).
For car owners, it saves money — and sometime makes it. For utilities trying to reduce fossil-fuel use, particularly during periods when electricity is in high demand, it’s a way of harnessing an expanding network of vehicle batteries to create virtual power plants that automatically spring into action when needed.
Toronto-based Kelvin Thermal Energy is pursuing similar AI capability for big industrial customers that want to store cheap electricity as heat inside graphite blocks, and then extract that energy when it’s most needed.
“It’s important that we have the ability to take, and stop taking, electricity when it’s most advantageous for the customer, so we’re leveraging all the great work that’s going on in AI and predictive analytics to achieve that,” says Stephen Roloff, the company’s Chief Strategy Officer.
Ecobee, for its part, is just getting started.
Founder and CEO Stuart Lombard has big plans for the rich pool of data his company’s smart thermostats collect.
Within the last year, he says, ecobee has hired nearly a dozen employees with PhDs in machine learning to explore new applications: “The fight for talent is big, but we’ll continue to grow that team over time.”
One service ecobee could soon offer, likely in partnership with utilities, is remote home-energy audits. Using AI to analyze thermostat data, the company can look at patterns and deduce what a homeowner can do to lower energy use. “Our goal would be to get within 10 per cent of the accuracy of in-home audits,” says Lombard.
Likewise, ecobee can help utilities tailor government incentives such as appliance rebates to the needs of specific customers, resulting in a more targeted and ultimately cost-effective use of scarce public dollars.
Even predictive maintenance is on ecobee’s radar.
The company is using machine learning to anticipate when an air-conditioner or furnace is likely to fail. If the equipment strays from normal operation, a home’s smart thermostat could dispatch an email or text to the homeowner recommending immediate maintenance.
“You don’t want to wait for your air-conditioner to break in the middle of a heatwave,” Lombard says.
The risk, of course, is that we put too much trust in the decision-making capability of this increasingly intelligent gadgetry.
An underlying machine-learning algorithm might be functionally sound, but its effectiveness depends on the quality of data being used to train it. That may not be a big problem for a smart thermostat that can be easily replaced if it doesn’t strike the right comfort-savings balance. But if AI is relied on to juggle supply and demand on critical infrastructure like the power grid, bad data will lead to dropped balls, whether that means damage to expensive equipment or community-crippling blackouts.
“Artificial Intelligence techniques draw conclusions from large masses of data, which may or may not include garbage data,” explains analyst Reinoud Kaasschieter of Capgemini, the global technology consultancy. “At a certain moment in time, it becomes impossible to determine on which data elements these predictions are based. In this way, artificial intelligence becomes black-box technology.”
The more we come to depend on AI, the more important it will be to test the veracity of the data being used.
“Like a curator in a museum who establishes if an exhibit is genuine or fake, a data curator should do the same for his data,” Kaasschieter adds.
Moving from electrons to molecules, this is something the petroleum industry will need to consider as it turns to AI for solving big problems.
If you’re the CEO of an oil company, for example, pipeline breaks are bad — even worse when they happen in the middle of populated or environmentally sensitive areas. It’s why more providers of pipeline monitoring services are beginning to tap the power of machine learning to improve accuracy and reliability.
Calgary-based Ingu Solutions Inc. has developed a golf-ball-sized sensor that it inserts into an operating pipeline. As it’s carried through the pipe, the sensor gathers a tremendous amount of data and can detect potential problems such as corrosion, cracks and the buildup of sediment.
“All of these anomalies or problems have unique data signatures. Every leak is different. Every case of sediment buildup is different. So what we need to do is sift through terabytes of data and scan for a signature that might represent a risk,” explains Ingu co-founder and CEO John van Pol.
No human, he adds, can do the task efficiently or effectively, which is why the company is experimenting with AI algorithms — using platforms such as IBM Watson and Amazon Web Services — to identify and flag signatures in the noise of all that data.
However, like the average homeowner or power-grid operator, Ingu and its customers crave dependability.
“What’s extremely important is that we avoid false-positives because, if we tell a customer there’s a leak, they will dig up the pipe to do a manual inspection or repair. If they dig it up and nothing is wrong, they’ll never use our technology again,” says van Pol. “So reliability, for us, is the most important thing.”
Thinking much like the Ontario government with its smart-thermostat giveaway, he adds: “AI is definitely something we need.”
Based in Burnaby, B.C., the company aims to build the world’s first commercial nuclear fusion reactor, and is using AI to accelerate development plans.
A major challenge is understanding how to create a controlled fusion reaction with plasma fuel. That means running simulations of how plasma behaves under a seemingly infinite combination of operating conditions and variables. Here, AI algorithms are being used to speed up analysis of simulations and even reduce the number of simulations required.
Chief Technology Officer Michael Delage says it allows General Fusion to optimize its processes and reactor settings faster than ever. “These tools are just critical if we’re going to take advantage of all the data we’re collecting.”
With headquarters in Toronto, energyX Solutions has developed an online, fully automated service called MyEnergyXpert, which lets home and business owners assess how they use energy and learn about the actions they can take to lower energy bills.
The company says its digital scan, which takes only 15 minutes, is just as accurate as those done by engineering firms but at one-tenth the cost. On average, it identifies improvements, such as installing new energy-efficient windows or replacing old appliances, capable of saving 25 per cent of current energy consumption for customers.
Co-founder and Chief Technology Officer Alex Corneglio says the next step is to enhance the accuracy of the service. Using machine learning, he says, the digital tool will do more than recommend actions for customers — it will reliably make decisions for them.
Machine-learning algorithms help Thermo.AI create the conditions in a power plant required for complete combustion of such fuels as coal or natural gas. Based in New York, the company employs special sensors to analyze moisture content in fuel, atmospheric pressure, interaction with air and other factors, making adjustments as necessary to assure the perfect burn.
AI helps customers get more energy out of their fuel, which not only reduces carbon emissions but extends the life and lowers maintenance costs of equipment by reducing wear and tear. Company co-founder Carolina Chaves Gonalez says the best way to make a huge impact is to make our current energy infrastructure more efficient.