At one point during the winter storm in December 2022, 60% of the US population was under winter weather advisories as the storm set records for 24-hour temperature plunges. |
As North Americans fought snow, high winds and plunging temperatures to gather for holiday celebrations at the end of December 2022, grid operators realized electricity consumption would surpass what the industry's most sophisticated software programs had predicted. And as suppliers scrambled to fire up more power plants, electricity prices soared.
Although no gridwide disasters occurred from the late December 2022 storm, dubbed Elliott, forecasts meant to predict electricity demand under the storm's extreme conditions had error rates of up to 9%, compared to the usual grid operator target of no more than 3%. The unexpectedly wide variances demonstrate how forecasting is becoming more difficult in an era of more variable weather conditions, economywide electrification and the transition to intermittent renewable energy resources.
"As we get more and more bad weather out on the extremes, the accuracy of forecasts is terrible in terms of the operational horizon, both day-ahead and week-ahead," said Alison Silverstein, a consultant who has previously worked at both the Public Utility Commission of Texas and Federal Energy Regulatory Commission. "Are there any industry standards? The answer is no. There should be."
Greater forecasting errors can bring grid instability, price volatility and human suffering. Grid operators typically use the mean, or average, absolute percentage error when comparing forecast-versus-actual electricity load. But Silverstein called this type of evaluation "wholly inappropriate" because "what matters is not the mean but the misses — and how big are we missing?"
In the wake of the December 2022 storm, FERC and the North American Electric Reliability Corp., the nation's primary reliability standards body, announced a joint investigation into the performance of the bulk power system, with a report scheduled for release in November.
RELATED: Holiday 2022 winter storm raises reliability, generation diversity questions
"As we witnessed in Winter Storm Uri, Winter Storm Elliott, and other extreme weather events, the demand for energy turned out to be far greater than had been forecast," former FERC Chair Richard Glick said in an email. "Although there are a number of reasons the regions affected by these events were challenged to keep the lights on, it is imperative that NERC, FERC, the [regional transmission organizations and independent system operators], and individual utilities consider the importance of improved load forecasting given the larger role extreme weather is having on grid reliability."
FERC spokesperson Celeste Miller said in an email that the agency does not have any proceedings open to examine the issue of forecasting electricity load.
Software snag
With severe weather moving toward Texas, the grid operator, the Electric Reliability Council of Texas Inc., on Dec. 21, 2022, forecast peak demand the next day would reach 67,775 MW. Actual demand on Dec. 22, 2022, was 8% higher, or 73,658 MW, according to an ERCOT review of the storm released March 27.
On Dec. 23, 2022, ERCOT set an all-time winter peak demand record of 73,910 MW, compared to a day-ahead forecast that was 4% lower.
Contributing to the variances was a software upgrade to ERCOT's primary forecasting system by Itron Inc. that went live Dec. 6, 2022. The upgrade resulted in forecasts double-counting the impact of holidays, when consumers typically use less electricity than normal, according to an ERCOT spokesperson, who added that the issue has since been resolved.
ERCOT's review said officials discovered the anomaly in the days leading up to the storm.
"This was not an issue with the software," Frank Monforte, Itron's director of forecasting solutions, said. Itron's software is used to forecast approximately 65% of the peak electricity load in the nine North American grid operators. "Our understanding is that because ERCOT hadn't had a chance to thoroughly test the enhancements, they chose not to use Itron's forecasts during Winter Storm Elliott and, instead, relied on other forecasts," Monforte added.
The enhancements included improvements in extreme weather forecasting and the ability to apply certain error corrections and select from any available model of raw weather data, Calvin Opheim, an ERCOT forecaster, wrote in a Feb. 16 presentation to ERCOT's Operations Working Group.
ERCOT's General Electric Co. backup system had models limited to a single weather forecast provider, according to Opheim, resulting in "higher forecast errors than if the Itron models had been operational."
A GE spokesperson said in a statement the company "is confident grid operators around the world can depend on our software to deliver accurate load forecasts, even in extreme weather events" and that GE is assisting ERCOT to ensure its components are configured properly ahead of future weather events.
ERCOT intends to seek bids for an additional vendor to supply a backup, externally generated load forecast, according to a spokesperson for the grid operator, adding that ERCOT's "conservative approach" means the day-ahead forecast is usually higher than the actual load.
"RTOs typically overestimate demand because this will tend to produce a system with more resources committed to producing energy on a day-ahead basis," Travis Kavulla, vice president of regulatory affairs for Houston-headquartered power producer NRG Energy Inc., said in an email. Officials in other markets are considering "whether to introduce market-based responses to this load bias."
"Typically, overestimates may overcommit resources that are not needed, and underestimates will pass up resources with longer startup times that would have been relatively economically useful," Kavulla, a past president of the National Association of Regulatory Utility Commissioners, said.
A man attempts to clear his roof of snow on Dec. 26, 2022, in Buffalo, NY. A winter storm dumped up to 4 feet of snow on the area, leaving thousands without power. |
Unprecedented weather
Very little about the December 2022 winter storm was normal.
Dan Woodfin, ERCOT's vice president of system operations, said in a Feb. 28 board presentation that the December 2022 cold weather came "deeper and quicker than the national weather models were forecasting."
The Southwest Power Pool had a 9% error rate in its winter storm peak load forecasts, higher than normal, spokesperson Meghan Sever said in an email. On Dec. 22, 2022, the SPP set a winter peak demand record of 47,157 MW.
"The presence of extreme wind chill without adequate historical data impacted our ability to determine its impact on load," Sever said. "[I]t's possible SPP's load forecast models may have adjusted to a lower demand with the expectation of snow and ice that we did not experience."
SPP uses GE, Enverus Inc. and internal load-forecasting systems, Sever said.
Peak load of roughly 140,500 MW on Dec. 23, 2022, in the PJM Interconnection LLC came in 8% higher than a forecast of 127,000 MW, a spokesperson said, and demand on Dec. 24, 2022, was 9% higher than forecast. A temperature drop of 29 degrees F over 12 hours in some areas Dec. 23, 2022, surpassed the previous PJM record of a 22-degree drop during a 2014 polar vortex.
Untrained algorithms
Such abnormal conditions have exposed a weakness in load-forecasting technology: It relies heavily on historical weather patterns.
"The load forecast is determined by algorithm, and the model had never seen the conditions that occurred on Dec. 23, with the confluence of unprecedented cold temperature drops, the holiday and the weekend," PJM said in a statement. "In some areas of the region, the difference between the high and low temperature on Friday the 23rd was one of the greatest in recorded history."
The PJM Interconnection underestimated peak electricity demand on Dec. 23, 2022, by 8%. |
Peak load in the Midcontinent ISO territory reached 105,916 MW on Dec. 23, 2022, compared with a day-ahead forecast predicting a peak of 100,033 MW. A January report by MISO's reliability subcommittee called the 5.5% error "abnormally high."
"The current industry practice for load forecasting is heavily dependent on historical data to train the forecast model," MISO spokesperson Brandon Morris said in an email. "Looking back at the historical data for the month of December, there were zero days that came close to matching the expected temperatures for Dec. 23."
ISO-NE spokesperson Matt Kakley said that while he cannot "point to something specific" to explain how the grid operator hit the bullseye, New Englanders are less dependent on electricity for heating. Though Christmas Eve was cold, "our temperatures may not have [been] as far off from normal as [other] regions, giving our modeling software more historical data to draw from."
ISO-NE did have more trouble forecasting electricity use during off-peak hours. The storm's Arctic air mass moved further east than had been projected, bringing colder-than-forecast temperatures, Kakley noted.
In response to a letter by 11 US senators expressing concern about ISO-NE load forecast inaccuracies, particularly in winter, ISO-NE President and CEO Gordon van Welie said Feb. 10 that the grid operator is expanding the number of cities modeled in weather forecasts, among other initiatives.
"Inaccuracies, even if small on a percentage basis, can turn into larger system-wide challenges when compounded by other events," the senators wrote.
S&P Global Commodity Insights produces content for distribution on S&P Capital IQ Pro.