Optimizing an ever-shifting electrical grid is becoming increasingly complex, especially as renewables are getting built more quickly than expected.
But it’s not just utilities that are puzzling their way through the transition. Energy trading is also becoming more complicated on the shifting grid. In general, traditional methods of forecasting are limited in their ability to predict pricing — and thus to maximize revenues for renewables projects.
That’s where artificial intelligence comes in. It’s still early days, but both legacy power players and startups are turning to the technology, in the hopes that it can forecast accurate prices for renewable energy and ultimately arm traders with better information in day-ahead markets.
Why is predicting the price for renewables so much more challenging than for fossil fuels?
In a word, uncertainty. Traditional forecasting models typically can’t handle the variability (or the fluctuating prices) that come on grids saturated with renewable energy, in part because of the huge quantity of additional data required.
Renewable electricity delivery is dependent on factors like geographic region, wind speeds, and weather patterns. Changing policy, regulation, and extreme weather add further complications.
Puneeth Kalavase, vice president of data science and engineering at Fluence, described the job of energy trading in the 90s and early 2000s as “kind of vanilla.”
“The prices of fuel didn’t change very often,” he said. ”Natural gas prices were stable over the weeks or months or even years timeframe; coal prices didn’t fluctuate that much.”
Today’s trading landscape is very different. There’s a mountain of additional information and variability, which means trading strategies have to change as rapidly as the grid itself. Batteries also complicate matters further, because of the very flexibility that makes them so useful.
“There’s no other asset on the grid that can both act as a load and a generator,” Kalavase added.
How could AI help?
Accurately predicting renewable energy prices requires processing large quantities of data from a wide range of sources, which is something AI tends to do better than the average human.
Using an algorithmic model to process temperature and wind speed data on any given day (or hour, or five minute window) is faster and easier than traditional methods. Fluence offers a confidence level alongside its price predictions, Kalavase said, which essentially represents how confident the software is that the price will be within a certain range.
But when it comes to actually recommending bids for customers, AI-powered price forecasts are just one piece. Fluence feeds price predictions through a non-AI mathematical optimization model aimed at finding the best outcome. The result is a recommendation of how much energy to buy or sell throughout the day.
AI can also account for the human shortcomings that can make any kind of trading challenging, said Lauren Kuntz, CEO and cofounder of the startup bidding recommendation platform Gaiascope.
“It’s up to our clients to take or leave our recommendations as they wish,” Kuntz said, though usually traders that stray from the recommendations fare worse. “Humans are emotional. Generally, you would have been better off just not stepping in.”
Where are we with implementation?
It will take a while for these technologies to see widespread adoption. One major challenge is that AI predictions require massive amounts of data, and it’s not always easily accessible in the power industry, or in a format that’s optimized for training models.
AI price predictions are not entirely untested, though. A variety of tools are already on the market, including GE’s Alpha Trader, which uses AI to predict energy generation and pricing and pair it with a portfolio’s specific risk profile. Meanwhile, Fluence’s bidding software Mosaic uses the technology to develop optimized bids based on forecasted prices, and Tesla’s Autobidder focuses specifically on battery assets, including price and load forecasting.
According to GE, Alpha Trader can increase trade values by $3 per megawatt-hour, while Fluence claims that using Mosaic can increase project revenue by up to 10% for renewables, and up to 50% for energy storage products.
Gaiascope, for its part, said it forecasts more than 3.8 million data points on the grid daily, and that its bid optimization and forecasting software has the potential to increase returns for clean energy generators by over 100%. And according to Rohan Ma, Tesla’s director of energy software and market participation, the company’s algorithms have yielded trading profits of over $330 million for storage investors.
Energy trading is one of dozens of applications for AI across the power sector. For a deep dive on other use cases, see our overview of how utilities are implementing it.
