To me, forecasting is a bit like that “pay no attention to the man behind the curtain” moment in “The Wizard of Oz.” After sitting through the “Integrating Energy Forecasts into Utility Planning and Operations: Perspectives from Market Participants” session at IEEE PES T&D in Chicago this April, I have to say it’s still rather murky to me. Thank goodness the forecasters are smarter than I am.
But, that’s why they forecast and I don’t.
I was personally intrigued by the opening sentence of the session description: “Energy forecasting plays a vital role in the planning and operation practices of modern utilities.” That made me willing to forgo lunch to sit in and listen.
And, I learned a bit. Most importantly, I learned that those people doing all that forecasting have a super important gig—both for maintaining the system and maintaining some happy finances for the utility as well.
T. Hong, professor at the University of North Carolina Charlotte, who teaches energy and analytics noted that “utilities have been forecasting since the inception of the power system. They’ve been doing this for over a hundred years now.”
A hundred years ago, they counted light bulbs, figuring one bulb for every house to give them potential load. Hong added that he taught a recent workshop in California where one attendee admitted that the “light bulb planning” system is still essentially functioning the same way.
And accurate profiles are getting more important these days—beyond light bubs, Hong noted. Market restructuring, large-scale integration of variable generation and grid modernization all impact power dynamics, and “all these uncertainties present huge challenges.”
Arnie De Castro with SAS had a challenge of his own to the audience: For short-term forecasts, how good do you think your forecasts should be?
To answer that question, first you have to think about which forecast. Utilities have weekly, daily and even hourly forecasts. And, after the fact, the actual loads are often pushed backed into the system to update future forecasts. But, even with the best historical data, there are other factors at play.
“More and more we’re getting distributed and renewable generation in our systems, and that impacts operation. CAL ISO is getting nearly 33 percent from renewables. That is huge in terms of trying to schedule a system,” he said.
But, returning to that question about how good your forecast should be, De Castro revealed that it only takes one or two percent to have a large impact.
In his example, using a small utility size (about the size of the City of Nashville), if the forecaster needed as little as an extra 1 percent, if would impact the system and leave the utility short, and a forecast missing one percent or two percent could cost thousands by running units they don’t need to be running.
Yes, as little as one or two percent.
“Energy and demand forecasts, even within two percent, have impact on costs and can affect proper operation of the distribution system—and can impact distribution operations with forecast of both magnitude and timing of peak demand,” De Castro added.
Scott Smith with Integral Analytics presented a PG&E case study that was more focused on long-term forecasting. PG&E’s goals included integrating the forecasting process among corporate, transmission and distribution levels; improving management, consistency and use of input data; introducing enhanced review and approval process supported by appropriate documentation to support regulatory assessment.
“As the circuit becomes more dominated by residential or commercial or more solar panels or not—you need to know the evolution,” Smith said.
That evolution may include historical data, but it may also include building out based on economic or social data as well. And, all of that is now a part of the forecasting umbrella.
Unfortunately, the math of forecasting requires consistency, but the reality doesn’t always parallel, especially with items like urban sprawl.
“What’s important is that you may not have the urban sprawl consistently all the time,” he added.
With this forecasting, they started out with regional objectives and estimated models for each applicable customer class: residential, commercial, industrial and agricultural. Then, they added in components for variables from historical data and economic models.
“Each model was tested to identify best the structure in terms of weather and economic variable, weather only, economic variable only, or just a constant term,” he added.
They do things similarly in North Carolina.
“What you’re looking at with load is human behavior,” said Jason Wilson with North Carolina Electric Membership Corporation. “When people get up. When people go to work.”
At NCEMC, they look at long-term forecasting, bringing in business needs and challenges from power supply, system planning, risk management, portfolio analysis, data quality, weather diversity, time constraints and regulatory requirements.
The co-op has moved from using monthly data to using hourly data with past implementations, but they have a desire to improve that. Currently, they are utilizing methodologies such as multiple linear regression, forecast trends drivers with economic data (GDP and households) and paying attention to patterns within that data.
“Data quality and finding trends are the kind of things that must be looked at personally. No matter how sophisticated your model, find patterns in the data that don’t look right and examine those,” he said.
“Ultimately, forecasting is a never-ending process of improvement,” he added.
To get your own peek at the people behind those forecasting curtains, check out the global energy forecasting competition, where they are leveraging the knowledge of the crowd and creating a community to tackle the forecasting challenge. Just click this link: www.gefcom.org.