By Michael Hainsworth
Keeping the lights on sustainably in today’s modern world requires Real AI. Not only do power producers need to maintain a constant stream of electricity, but they contend with issues that Sir Adam Beck never could have expected more than 100 years ago when he founded what is known today as Ontario Power Generation: customers feeding-back their own electricity from solar panels, and heavy industry charging backup batteries in the wee hours of the morning when power is cheap.
“You’ve got two sets of generation and variable demand in the middle and that’s all interacting,” points out Nick Pender, the Vice President of Energy Markets at OPG. “And that’s the inherent complication of the system. And it changes, you know, every five minutes.”
Pender is using artificial intelligence and machine learning to help determine changes in cost of generating electricity, and when it makes sense to sell energy to neighbouring provinces like Quebec and Manitoba, and into regions still using coal-fired plants like Michigan. “The degree to which we can trade with our neighbours and do that economically, I think, is beneficial for society,” says Pender, pointing out, “emissions don’t care about borders.”
What Pender and his team of energy traders is doing with AI is moving beyond the classical rules of cause and effect. By unbounding that process and taking a lot more information, Pender can look for patterns that probably didn’t exist before.
From cellphones to electric car chargers, grid operators are witnessing hundreds of thousands of people doing similar activities at slightly different times, and that changes the profile of demand. “And the really interesting bit about power… is that you just can’t store it. You can put it in batteries and you can delay it for a little bit, but not to the same degree as a harvestable crop. It’s real time. So that demand, fluctuation and change, has to be managed in real time.”
The machine learning algorithms Pender is harnessing pulls in data from a variety of sources, including the weather. This is particularly critical for understanding Canada’s largest solar farm in Kingston, Ontario. It generates more than 100 megawatts of renewable energy through more than 460,000 panels. That’s enough to power 17,000 homes – although not constantly. There is even more solar available when we consider neighbouring residential rooftops. “Of course we’ve had clouds since forever, but now the system is more complicated than it was,” he laughs. “If a cloud goes over Kingston load can change hundreds of MW in minutes… so thinking about how cloud cover and how wind changes the supply and demand of power is a really critical feature.”
In interconnected markets, Real AI is observing wind and solar time-of-day behaviour, but also the slightly different drivers for demand in these regions. “So, perhaps people eat dinner at slightly different times in Minnesota, or it’s hotter in New York. There are market signals happening at our borders. We have to compute all that together and try to work out what’s going to be happening in the next period of time.”
“Overnight we would expect demand to be low because a lot of people are home and sleeping. Businesses are open, but not as much as in the day,” says Pender, adding that the AI sometimes spits out surprising insights. “What we started to see is batteries start to charge at night.” It’s led to competing AIs as heavy industry sees a cheap opportunity to fill up its batteries. “You can have a smooth demand profile, and all of a sudden, peak up for maybe a half an hour. This is, for me, where the AI is really helpful.”
Does Pender and OPG feel like AI is the right direction, empowering his team and maintaining more control? “100%.”
