Power forecasters struggle to find grid’s ‘new normal’

By Jeffrey Tomich | 04/14/2020 07:05 AM EDT

From a soaring unemployment rate to a roller-coaster ride for stock markets, economic vital signs are showing the strain of an unprecedented pandemic that is changing virtually every aspect of American life.

Grid operators monitoring electricity use in a control room in PJM Interconnection's headquarters near Valley Forge, Pa.

Grid operators monitoring electricity use in a control room in PJM Interconnection's headquarters near Valley Forge, Pa. PJM Interconnection

From a soaring unemployment rate to a roller-coaster ride for stock markets, economic vital signs are showing the strain of an unprecedented pandemic that is changing virtually every aspect of American life.

Energy analysts and economists are still trying to wrap their arms around how the coronavirus has affected the power sector over the past three weeks. But another small segment of the industry has a more daunting task: predicting the future.

They are the forecasters — analysts, engineers and meteorologists tasked with predicting power use for every hour, every day. They work for grid operators, utilities and private firms. And their job is at the heart of ensuring the lights stay on at the lowest cost possible.


Unlike supermarkets or department stores that can store goods for days, weeks and even months, the grid is unique because electricity supply and demand must constantly be in balance.

At worst, significantly underestimating power consumption at a given hour could leave system operators scrambling to bring plants online to keep the grid stable and avoid power disruptions. But even overestimating load can mean higher costs that ultimately trickle down to consumer bills.

"If we make the commitment for [plants] to be online to manage reliability and then it turns out we don’t need them, they need to be made whole," said J.T. Smith, director of operations planning for the Midcontinent Independent System Operator (MISO), which oversees the bulk power grid across a wide swath of the nation’s midsection, from the Gulf Coast to Canada. "It’s not their fault they’re online. It’s the call of the operators. That is then a cost that gets spread across the footprint."

Forecasters have dealt before with economic downturns. There was the Great Recession a decade ago. There have been natural disasters, from Gulf Coast hurricanes to Superstorm Sandy and California wildfires. But never has there been such a broad, deep and erratic slide in electricity use as there has been in recent weeks while millions of Americans stay sequestered in their homes and as schools, restaurants and offices remain dark.

"It seems like every time we learn what the new normal is, something else changes," Smith said.

The past three weeks have brought a patchwork of stay-at-home orders in states across the country. Social distancing measures have been implemented by some cities but shunned by others. Weekday usage patterns have shifted as people plug in to work from home.

"It’s a cat-and-mouse game," said Blagoy Borissov, forecasting manager for MISO. "The uncertainty that we’re fighting with is trying to find out when is the load going to change, by how much and what is that attributable to?"

Not only has demand eroded, peaks are occurring at different times of the day. Weekday peaks that used to occur at 8 a.m. are now nearer to lunchtime, and evening high points — correlating with people returning home and turning on lighting, appliances and air conditioning — are no more.

"We have not seen this level of continuous load changing day after day," Borissov said in an interview.

No reference point

Grid operators such as MISO and PJM Interconnection LLC haven’t seen any reliability impacts due to COVID-19. But in briefings on the pandemic’s effect on grid operations, officials acknowledge load forecast errors have increased.

Part of the challenge is visibility.

"We see things aggregated at a higher level," said Chris Pilong, director of operations planning at PJM, which manages power flows across 13 Midwest and Mid-Atlantic states and the District of Columbia. "As things shift, from people being in offices to being at home, all we see is the net difference.

"Trying to predict what the demand is going to be has a little bit of a psychology component to it. What are people doing? You don’t have perfect foresight, but we use some general indicators as far as what’s going to drive people’s behavior."

Aidan Tuohy, program manager for the Electric Power Research Institute, said forecast models rely on machines learning algorithms to crunch historical data and look for trends and patterns that can help predict electricity demand.

The challenge with the recent drop in electricity demand is the unprecedented nature of how it has unfolded.

"These models do catch up pretty quickly," Tuohy said. "They continually look at weeks and days and weigh themselves to what’s happening."

There have been repeated attempts in recent weeks of comparing weekday electricity demand to a weekend, holiday or even a snow day, when kids are home from school and some parents are working at home on laptops.

Rob Allerman, senior director of power analytics at Enverus — a Houston-based energy analytics company that sells forecasts to utilities, traders and big energy users — said the reality is there are no good analogies for what’s happening.

"We’re seeing something that we’ve never seen," he said.

Indeed, the unprecedented nature of the coronavirus pandemic, its effect on the economy and changes in electricity use pose thorny problems for forecasters who rely heavily on historical data and algorithms.

Allerman, who was previously head of North American power analytics at EDF Trading and worked in power analytics at Deutsche Bank, runs models using four different temperature forecasts and automatically weighs them based on their accuracy over a given time period.

"What the [model] has been seeing is that 50,000 megawatts at 8 a.m. is now 45,000 MW, so it learns that one day, and it will make that adjustment downward. If it sees that the next day, it will further adjust and keep adjusting and learning," he said.

In general, the biggest shifts in forecasts have correlated with stay-at-home orders.

Among the nation’s big power grids, Allerman said New York and California — areas that were among the first to adopt shelter-in-place requirements — saw the most immediate demand destruction. The downward spiral came slightly later in MISO and PJM, and the Southwest Power Pool and the Electric Reliability Council of Texas.

In addition to trying to predict what hourly demand will be, forecasters at grid operators and companies like Enverus are crunching numbers to isolate the effect of COVID-19 on power demand and determine what it would have been under normal circumstances.

An even greater challenge may lie in the weeks and months ahead.

Whenever stay-at-home orders are eventually lifted, demand will pick up as many stores and restaurants reopen and people return to work. Idled factories will resume production. And if that occurs later this spring or during summer, the return to normal will coincide with the usual pickup in electricity demand tied to air conditioning load.

Tuohy said electricity demand lost to the coronavirus shutdown is likely to return at a slower rate than it left.

It’s something grid operators like MISO and PJM are also thinking about — as usual with a cautious eye.

"Right now, on the way down, we’ve been overforecasting," MISO’s Smith said. "There is a concern that as we start going the other direction, are we going to start running any reliability risks on underforecasting what future load will be?

"It is something we are actively paying attention to so that we don’t put ourselves at risk."