Many applications of artificial intelligence for the grid are still in early, even theoretical stages. Energy forecasting, though, is one area where the market for AI is more mature — even utilities, which tend to be cautious with new technologies, have been experimenting with the tools for years.
Hitachi Energy’s new AI-driven forecasting solution is the latest forecasting tool to hit the market. Called Nostradamus AI, it uses AI to provide forecasts on load, market pricing, and renewable energy generation that the company says can be more than 20% more accurate than some industry targets. The cloud-based solution is targeted at customers such as utilities, renewables operators, or energy traders.
Bret Toplyn, director of product management at Hitachi Energy, told Latitude Media that one of today’s main challenges for energy forecasters is the high influx of data — so high, in fact, that it’s now beyond the processing capabilities of traditional forecasting methods.
The data required to train energy forecasting models is constantly evolving, and the accuracy of the forecast they produce does not necessarily improve the more historical data you feed them. That’s a different challenge from large language models, which tend to become better the more information they receive.
Another hurdle is the sheer speed of digital transformation, and the need to be flexible to change along with it.
“The challenge is to build something that can continue to evolve with the industry,” he said. “And that’s one of the main pieces that we factored into the design of the product so that it’s highly flexible, highly scalable, and very dynamic.”
To achieve the needed flexibility, Nostradamus AI builds each forecasting “pipeline” using code blocks that can be tied together in different combinations to produce the desired forecast. This modularity allows developers to save time by reusing existing code blocks, and focus their attention on creating new code blocks that keep up with the constant changes in the energy market.
“For example, we might have a market price pipeline for forecasting market prices,” Toplyn said. “We can easily use and reuse some of those code blocks, and then build new code blocks for an entirely new pipeline to predict the transformer failure for heavy assets.”
Paired with both continuous learning and an algorithm-agnostic approach — which allows Hitachi Energy to leverage a variety of open source or proprietary algorithms to do the forecasting — the tool should be to keep up with the constantly evolving market. If there’s a policy change that affects the market, for example, the developers would be able to introduce new variables and “conduct a retraining” to shift the view of the model going forward.
1% accuracy can mean millions of dollars
And better forecasting accuracy offers a huge payoff for utilities, grid operators, and others responsible for the grid’s smooth delivery of energy..
“Even a 1% increase in the accuracy can translate to millions of dollars, just because of how significant the operations are in the electrical grid,” Toplyn said. “That’s valuable not just to the companies owning the assets or meeting the demand, but to the grid in general, because everything can theoretically operate more efficiently.”
Hitachi Energy claims Nostradamus AI can generate better forecasts than some industry targets, for both load and price. While load forecasting has an industry target accuracy in the high-90% range, price forecasting falls closer to the 70-80%, Toplyn said; Nostradamus AI, he added, achieved price forecasts in the 90% accuracy range.
As of this month, the tool is available to anyone interested in shelling out for a subscription. One global transport and logistics company headquartered in Europe, which Hitachi Energy didn’t name because the details aren’t yet public, has already been using the tool on a pilot project to automate and predict bidding behavior for battery assets.
Going forward, Hitachi Energy anticipates “significant demand.” The company is working to add new features to the product, such as giving customers the ability to pull from a day in history that resembled what they’re expecting, as well as probabilistic forecasts, which are a range rather than a single point.
Toplyn said that today’s twin challenges — of the huge amount of data that’s now being collected and the rapidly changing market — is a combination that “lends itself very well to artificial intelligence.”
“Having a tool specifically built around that influx of data and focused on the evolution of the industry allows us to reach an entirely new group of users who may have been aware of forecasting but haven’t necessarily taken the leap for it yet,” he added.
