185. AI-Powered Energy Forecasting: How Accurate Predictions Could Save Your Power Company
The POWER PodcastShow Notes
Net-demand energy forecasts are critical for competitive market participants, such as in the Electric Reliability Council of Texas (ERCOT) and similar markets, for several key reasons. For example, accurate forecasting helps predict when supply-demand imbalances will create price spikes or crashes, allowing traders and generators to optimize their bidding strategies. It’s also important for asset optimization. Power generators need to know when to commit resources to the market and at what price levels. Poor forecasting can lead to missed profit opportunities or operating assets when prices don’t cover costs. Fortunately, artificial intelligence (AI) is now capable of producing highly accurate forecasts from the growing amount of meter and weather data that is available. The complex and robust calculations performed by these machine-learning algorithms is well beyond what human analysts are capable of, making advance forecasting systems essential to utilities. Plus, they are increasingly valuable to independent power producers (IPPs) and other energy traders making decisions about their positions in the wholesale markets. Sean Kelly, co-founder and CEO of Amperon, a company that provides AI-powered forecasting solutions, said using an Excel spreadsheet as a forecasting tool was fine back in 2005 when he got started in the business as a power trader, but that type of system no longer works adequately today. “Now, we’re literally running at Amperon four to six models behind the scenes, with five different weather vendors that are running an ensemble each time,” Kelly said as a guest on The POWER Podcast. “So, as it gets more confusing, we’ve got to stay on top of that, and that’s where machine learning really kicks in.” The consequences of being ill-prepared can be dire. Having early and accurate forecasts can mean the difference between a business surviving or failing. Effects from Winter Storm Uri offer a case in point. Normally, ERCOT wholesale prices fluctuate from about $20/MWh to $50/MWh. During Winter Storm Uri (Feb. 13–17, 2021), ERCOT set the wholesale electricity price at its cap of $9,000/MWh due to extreme demand and widespread generation failures caused by the storm. This price remained in effect for approximately 4.5 days (108 hours). This 180-fold price increase had devastating financial impacts across the Texas electricity market. The financial fallout was severe. Several retail electricity providers went bankrupt, most notably Griddy Energy, which passed the wholesale prices directly to customers, resulting in some receiving bills of more than $10,000 for just a few days of power. “Our clients were very appreciative of the work we had at Amperon,” Kelly recalled. “We probably had a dozen or so clients at that time, and we told them on February 2 that this was coming,” he said. With that early warning, Kelly said Amperon’s clients were able to get out in front of the price swing and buy power at much lower rates. “Our forecasts go out 15 days, ERCOT’s forecasts only go out seven,” Kelly explained. “So, we told everyone, ‘Alert! Alert! This is coming!’ Dr. Mark Shipham, our in-house meteorologist, was screaming it from the rooftops. So, we had a lot of clients who bought $60 power per megawatt. So, think about buying 60s, and then your opportunity is 9,000. So, a lot of traders made money,” he said. “All LSEs—load serving entities—still got hit extremely bad, but they got hit a lot less bad,” Kelly continued. “I remember one client saying: ‘I bought power at 60, then I bought it at 90, then I bought it at 130, then I bought it at 250, because you kept telling me that load was going up and that this was getting bad.’ And they’re like, ‘That is the best expensive power I’ve ever bought. I was able to keep my company as a retail energy provider.’ And, so, those are just some of the ways that these forecasts are extremely helpful.”
More from The POWER Podcast
In this episode of The POWER Podcast, Sandeep Kumar Gupta, sales director with Siemens Energy Gas Services, and Naira Elsayed, a senior project manager with Elsewedy Electric Power Systems Projects (PSP), discuss the Rabigh Reinforcement Combined Cycle Power Plant, a finalist for a 2026 POWER Top Plant Award.
Utilities and independent power producers have spent years adding sensors to their generating fleets and creating dashboards to view what they relay. But visibility into how a plant is operating isn’t the same as knowing what to do about it—and that gap is where Siemens Energy’s Omnivise Energy Management System is increasingly shedding light, explained two of the company’s experts as guests on The POWER Podcast. “Monitoring is just telling you what’s happening,” said Sarah Dietrich-Smith, Siemens Energy’s commercialization manager for Omnivise Energy Market and Fleet Management Solutions, “and Energy Management is telling you what should happen next, and it helps you to execute that decision,” she explained. The distinction, she said, is the real difference between tracking a plant’s condition and actually optimizing it—moving “from just this visibility to true optimization that you can actually take action on, and optimization that’s based and tied to measurable business outcomes.” From Single Assets to Whole Fleets Katie Hanley, head of Siemens Energy’s Omnivise Energy Market and Fleet Management Solutions team, framed Energy Management as fundamentally a question of business outcomes. Given a plant’s generation commitments, gas hedging position, and access to additional electricity markets, “What are the optimized economics for running your plant or your fleet?” she asked. “Historically, assets were managed individually. They really only cared about the question, ‘How can we get the single asset to perform better?’ ” Dietrich-Smith said. “We want to move into a fleet optimization mindset.” One plant might be constrained while another has more flexibility—sometimes for regional reasons, such as differing prices or curtailment risk in different markets. “This fleet-level approach is what’s going to help operators allocate flexibility, manage this risk, and ultimately optimize the commercial performance across the full portfolio,” she said, “instead of just making that isolated asset or site-level decision.” Dietrich-Smith also pushed back on a common assumption about where the value comes from. “I think the most common misconception is that the only value we’re bringing is efficiency improvement,” she said. Customers also see fewer penalties, more effective participation in electricity markets, and faster response to changing conditions—benefits she summed up simply: “Time is money.” Renewables Complicate the Math A growing share of that fleet-level complexity traces back to renewables, Hanley said, whether a customer operates a hybrid portfolio directly or simply competes in a market where renewable penetration is rising. She described thinking in terms of dispatchable versus non-dispatchable generation: as more renewables come online, dispatchable options become scarcer, making it harder to maintain grid stability—and more important to predict when instability might occur. That prediction problem pulls in weather and market data alike, from cloud cover and wind speeds at different altitudes to real-time electricity prices. For a hybrid site with battery storage, the same data informs decisions about when to charge from onsite generation versus the grid, and when to discharge—sometimes purely to arbitrage low prices rather than to support the site’s own generation. “These batteries are creating dispatchable flexibility when normally we would only be able to do that with some of our large gas assets,” Hanley said, adding that customers are increasingly weighing a battery discharge against ramping up a plant or duct burning, based on which is the better economic call in the moment. Hanley closed the conversation on a practical note for listeners still weighing where to start. “Everything is only getting more complex, so give us a call,” she said.
Public power utilities have long promised lower bills, reliable service, and decisions made close to home—and the data still backs that up. But holding that ground is getting harder. In this episode of The POWER Podcast, Scott Corwin, president and CEO of the American Public Power Association (APPA), talks about the forces now testing the community-owned utility model. Corwin explains why resource adequacy, reliability, and affordability have converged into the defining challenge for APPA's roughly 2,000 member systems—a concern that once affected only certain regions but now stretches across nearly the entire membership. He breaks down the tools public power uses to keep rates down, including tax-exempt financing and elective pay credits, and why local control itself acts as a check on spending. The conversation also digs into the issues reshaping the sector: how smaller utilities are evaluating massive data center interconnection requests while protecting ratepayers; why Corwin describes public power's generation strategy as "any of the above" rather than "all of the above"; and the structural headwinds—permitting delays, an aging workforce, supply chain strain, and cybersecurity—that determine what utilities can build and how fast. Corwin closes with what he most wants policymakers to understand about community-owned utilities, and the value of 2,000 diverse communities speaking with one voice on affordability and reliability. Listen to the full conversation to hear how public power is navigating one of the most dynamic periods the industry has faced in years.
A perfect storm is brewing across the U.S. power generation fleet. Between 600 and 700 GE 7FA gas turbines installed during the bubble of 2000–2004—plus roughly 900 7EA units worldwide—are simultaneously approaching the OEM's 144,000-hour, 5,000-start end-of-life threshold. Superalloy forgings carry multi-year lead times. Only a handful of shops worldwide can service these machines. And the data center boom is pushing utilization higher than anyone planned for.
Duke Energy operates 11 nuclear units across six sites in the Carolinas — a fleet that produces more than half of the region's electricity year in and year out. In 2025, that fleet posted its best capacity factor on record, north of 97%.