Advancements in forecasting have improved the reliability of day-ahead and hour-ahead estimates of wind generation. Wind never will behave like a base-load power plant. But as system operators integrate wind forecasts into their planning and market processes, they’re transforming intermittent wind energy into a variable but reliable resource.
When transmission system operators think about wind power, two problems leap to mind. First, wind resources tend to be located far away from load centers, so new wind farms sprouting up across the countryside will require new long-distance transmission lines. Second, for purposes of system operations, wind is an “intermittent” source of power—unpredictable and non-dispatchable. As a result, operators have viewed wind power as an energy-only resource, essentially useless (and sometimes detrimental) for system reliability and security.
The first issue, the need to build transmission capacity to serve wind farms, is part of the hottest regulatory debate in the industry today. Namely, who should pay for new transmission capacity that arguably won’t benefit all customers equally, and how should it be regulated? (See, for example, Bruce Radford’s “Taking Green Private” March 2010, and “T Party Revolt,” October 2009). But while this argument has raged in FERC hearings and other forums, the second problem steadily has diminished in importance.
Advancements in wind forecasting in recent years have vastly improved the reliability of day-ahead and hour-ahead estimates of wind generation. And as system operators, utilities and energy traders integrate wind forecasts into their planning and operations, they’re transforming wind’s reputation. What system operators once considered a wholly unreliable resource, they increasingly view as a predictable and manageable source of emissions-free energy.
“The industry is moving away from using the word ‘intermittent’ and started using the word ‘variable,’” says Kevin Sullivan, a senior vice president with KEMA Consulting. “The word ‘variable’ implies control. That change in thinking has happened because the technology has improved to make wind energy more manageable.”
Also, this change might be attributable, at least in part, to the wind industry’s own public-relations efforts. For instance, the American Wind Energy Association advocacy group studiously avoids using the term “intermittent” because it implies complete unpredictability. However, even though forecasting technologies and operational practices are bringing wind out of the “intermittency” ghetto, the industry still has work to do before system operators will treat wind as a generating capacity resource.
“Wind forecasting helps, because the more you can reduce uncertainty around scheduling, the more you can fit wind generation into the system,” says Mark Ahlstrom, CEO of forecasting company WindLogics. “But you’ll never be able to lock it in a day ahead and treat it like a coal plant, because of the nature of the wind resource. Forecasting will never be perfect all the time.”
System operators don’t need it to be perfect; they need it to be highly predictable. And with recent and ongoing technology developments, that goal now appears to be within sight.
In mid-April 2010, stakeholders in the Midwest Independent Transmission System Operator (MISO) convened to discuss critical issues facing the ISO. The issue of integrating wind energy into the network ranked foremost on the agenda.
MISO’s focus on wind isn’t surprising, given the fast-growing role of wind for powering the network; wind capacity in MISO nearly doubled in 2009, going from 4,300 MW in 2008 to more than 7,600 MW today. And MISO counts more than 60,000 MW of proposed projects within its footprint. Accordingly, the MISO is making tariff and rule changes to bring wind fully into the fold.
“Right now wind farms are considered price takers,” says Michael McMullen, director of MISO’s West regional operations. “But we’re developing a ‘dispatchable intermittent’ feature that would allow wind farms to register as dispatchable generators. They’d be treated the same as traditional generators.”
Resources registered as “dispatchable intermittent” wouldn’t be counted on for regulating or contingency reserves, but from a market perspective they could offer and sell power into the MISO market on the same terms as other dispatchable generators—with one exception; the ISO would dispatch these resources up to the maximum megawatt-output figures provided by a real-time forecasting system, rather than the amount the generators offer to the market.
Real-time forecasting is the key. MISO has been using a centralized wind energy forecasting system since mid-2008. The system, provided by Energie & Meteo of Oldenburg, Germany, provides MISO operators with hourly wind forecast data for a seven-day period. McMullen says operators rely mostly on intra-day and day-ahead forecasts. “Getting forecasting data in house gives you a great big jump forward to see within an hour or two, when wind output will be ramping up or down,” he says.
Wind forecasting has obvious appeal for system operators, especially those in sprawling, wind-rich areas like the nearly 1-million square mile MISO region. “Very large balancing areas with adequate transmission take maximum advantage of diversity in both load and wind generation,” stated an exhaustive study of wind-integration potential in the Eastern Interconnection, published in January 2010 by engineering-consulting firm EnerNex, under contract with the DOE’s National Renewable Energy Laboratory (NREL).1 Size matters when it comes to centralized wind forecasting.
“Geographic diversity benefits the accuracy of an aggregated forecast,” McMullen says. “It means that if you happen to miss the timing of a wind ramp in one place, it probably is hitting someplace else. We recognize that as a benefit that gives us some nice accuracy numbers in current and next-day forecasting.”
Specifically, while state-of-the-art forecasting technologies for a specific plant might predict, on average, day-ahead wind output with error rates between 12 and 20 percent, aggregating forecasts across a whole region can cut those error rates in half or better, to around 5 percent (see Figure 1). Such numbers, calculated as a factor of installed capacity, might be a bit deceiving; on average, wind farms generate about 30 percent of their nameplate capacity. Nevertheless, with forecast error rates falling into the single digits, wind starts to more closely resemble the other major variable on the grid—shifting load, which operators forecast with relative error rates from about 1 to 4 percent.
“Bulk power system operators always make assumptions about load, based on experience with weather and load forecasts,” says Mark Lauby, director of reliability assessment and performance analysis for the North American Electric Reliability Corp. (NERC). “Wind forecasting is just another tool operators will use to ensure they have the flexibility they need to respond to changing conditions.”
Further, because weather phenomena influence both load and variable power sources, it makes sense for system operators to integrate the way they use forecast data. Indeed, NERC is considering recommendations2 that anticipate eventual convergence between load forecasts and generation forecasts for wind and other variable power sources, including solar and run-of-river hydro. In the short term, NERC’s new recommendations would make wind forecasting a standard part of control-room procedures and energy management systems (EMS). And earlier this year, the Federal Energy Regulatory Commission (FERC) issued a notice of inquiry seeking input on wind-integration issues, including several possible forecasting requirements that would affect wind generators, utilities, transmission owners and system operators.3 A large share of market participants and ISOs with substantial amounts of wind capacity already are using forecasts,4 but before now neither FERC nor NERC have defined it as standard procedure.
“Over time, as we gain experience with [wind forecasting] technology and learn its strengths and weaknesses, we’ll get more certainty around it,” Lauby says. “It’s the same with every new technology that comes onto the system, whether it’s gas turbines or nuclear plants. At first there’s some skepticism, but with experience we learn how to optimize it.”
Smart-grid and dynamic pricing programs are turning customer load into a more easily controllable factor, potentially making it a more important tool for system balancing. Utilities and system operators historically have tapped into demand-response mechanisms mostly to manage the critical peak, curtailing power for a few large electricity customers, such as factories and aluminum smelters, to prevent brownouts and blackouts from rolling through the network. But retail-level DR programs—which trim demand across entire communities—might give grid operators another tool for managing variables.
“On the left and right, these two variables can be used to balance each other,” says Don Leick, product management director with Telvent, which provides smart-grid systems and weather- and load-forecasting services. “Load-modeling software ties into an EMS. I don’t see anyone with a completely integrated solution yet, but these are pieces that can fit together.”
Eventually, if the system gets smart enough, the operators’ EMS might present variable generation and demand factors together for decision-support purposes—perhaps someday as virtual levers the operator might use to balance the system in a way that makes optimal use of current and forecasted resources.
“The smart grid and distributed demand response allows you to integrate large amounts of renewable energy,” says Pascal Storck, a vice president with renewable energy forecasting and resource assessment company 3Tier. “Forecasting allows distributed DR to be deployed and used more intelligently than if you didn’t have forecasting.”
But as with all new ideas, distributed DR might also bring unintended consequences. Changing behavior affecting customer demand might complicate load forecasting for system operators, who make ongoing decisions based on load rising and falling in a predictable way.
“Smart grids offer a host of reliability benefits, and could allow operators to see patterns they’re not seeing now,” Lauby says. “But to some degree, bulk-power system operators assume certain characteristics about load and the direction of energy flows, based on experience. Now, with two-way flows and with communication between customers and the market, the predictable nature of the load will change. We’ll need to get experience with that.”
As valuable as forecasting is becoming, system operators harbor well-justified fears about the inherently chaotic behavior of large weather patterns—and the way they can rapidly change the output from wind farms. Large-scale ramping events, in which the wind suddenly dies or rises to a gale, make it difficult for system operators to rely on wind capacity. And they create special problems for forecasters, because these events are atypical and tough to predict—precisely because they’re atypical.
“To get a good forecast, you need a good sample of events,” says John Zack, director of forecasting for AWS Truepower (formerly AWS Truewind). “Most wind farms haven’t been operating for a long time, so you’re always facing new and different events that you haven’t seen before in the history of the system. And the big events, the ones that are of greatest interest for reliability and security, are anomalous events.” As a result, forecasters have little or no data to show them the specific local conditions that lead to major ramp events at a given site.
“Predicting large-scale ramps is tricky, and it will remain tricky,” Leick says. “Forecasts can indicate a high probability that a thunderstorm will occur tomorrow afternoon, and that you’ll experience some ramp events. But will it be at 1:30 or 2:15? That’s the level of precision people want.”
In practical terms, gauging the exact timing and location of a large-scale ramp event is beyond the reach of current technologies and weather models. As a result, researchers are focusing a great deal of attention on this area.
Initial approaches are using information technology to produce more accurate and useful forecast information for operators. For example, probabilistic forecasts can provide a clearer idea of the mathematical chance a ramp event will happen within a given time period. “Operators need to know what the probability of an event is in order to assess the risk and decide whether to take action,” says Storck of 3Tier. So forecasting companies are working to refine the way their systems calculate and provide probabilistic data.
Other computational approaches include high-performance modeling systems, such as those developed by the Geophysical Fluid Dynamics Laboratory at the National Oceanic and Atmospheric Administration (NOAA). Although such systems are developed for weather forecasting generally—to help the National Weather Service (NWS) predict thunderstorms, tornadoes and blizzards—the same technology helps wind energy forecasters to improve their computational models of the conditions that lead to ramp events.
Ultimately, however, all models depend on the quality and quantity of data that feeds into them. And given the lack of historic data, what forecasters really need is a larger sample of real-time and ongoing wind measurements.
“Input data limits forecasting ability,” Zack says. “Observations available right now aren’t adequate to forecast ramp events even a few hours ahead. To get a big improvement you need better measurements of the atmosphere than we have right now, and you need different measurements than those that typically serve weather forecasts.”
Given NWS’s mission to predict weather events that threaten public safety, its work concentrates on understanding conditions where storms are spawned, in the upper and middle atmosphere. “There aren’t a lot of observations of winds near the earth, and those observations are critical for forecasting wind-ramp events,” Zack says.
Such observations come from wind-measuring equipment in the field, but there’s a practical limit to the number of anemometers that forecasting companies and wind developers can afford to install. AWS is trying to work around this limit, to some degree, by building forecast-sensitive models that might help researchers identify locations where anemometers can yield the most valuable measurements.
“How do we get the most out of our observational dollar?” Zack asks. “The answer is to do observational targeting, using models that point out areas that will be most sensitive for forecasting for a given site, depending on topography, time of year and other factors.”
AWS is proving out this approach in a DOE-funded project in Hawaii. “Hawaii has one of the most critical wind management problems of anywhere in the United States,” Zack says. “It’s an island grid with no regional interconnection, so they have no flexibility, and they have a very high wind penetration. That’s why DOE targeted that area to test methods for improving the short-term forecast.”
In addition to the Hawaii project and other efforts to advance forecasting technology, DOE and NOAA are considering options for large-scale deployment of measurement devices—including anemometers and also wind profilers. Profilers are sensitive Doppler radar stations that point upward to measure wind speed and direction at various elevations. Deploying such devices would be extremely helpful for improving the resolution of wind models, but at a cost that’s beyond what wind developers or forecasters can reasonably afford—hence the rationale for a government program that might cost billions of dollars.
Fortunately, such a wind-measurement Manhattan project might not be necessary to provide a major breakthrough in observational data.
“We need to develop technologies that bring costs down to a manageable level,” Zack says. “For example, we’re excited about low-cost Doppler radar systems that might be mounted to cell towers, and could give you good boundary-layer wind measurements over a 30- to 50- kilometer radius.” Such technologies are in early development phases, but by about 2015 they might become economical enough to allow private companies to deploy a whole network of them, Zack says.
“You could take a lot of measurements remotely and feed the data rapidly into forecasting models. That’s where the technology needs to advance to really improve forecasting.”
Over time, as wind forecasts improve and become more tightly integrated into the control room, system operators will become more comfortable relying on wind power to serve load requirements. Given the nature of the wind resource, and the critical importance of maintaining reliability and security, wind might never be used as firm capacity. But already forecasting is raising the value of wind generation, both in system operations and wholesale energy markets.
Energy traders, for example, use wind forecasting data in highly sophisticated and integrated ways to improve their ability to predict market prices. And utilities and wind facility owners use forecasting data—both site-specific and centralized models—to inform their operational and marketing decisions. “They’re making choices about whether to rely on wind megawatts on an hour-ahead or day-ahead basis, or to buy backup power,” says Michael Grundmeyer, a vice president with 3Tier. “They’re making long-term decisions about how much peaking capacity to buy or trade, and for making purchase decisions on natural gas and tolling options. They’re looking at how much wind will displace the fuel on the margin, because if the wind falls off, it has an effect on the price of fuel.”
In some organized markets, traders are using wind forecast data to help them decide whether to buy or sell financial transmission rights (FTR). “Forecasts help traders figure out where wind megawatts will land and at what time,” Grundmeyer says. “That helps them manage the transmission they can access.”
And ISOs, too, are looking to financial trading data about wind energy to help them better plan their system-balancing needs. “Intra-day capacity commitments in the market feed into our current and next-day operating assessments,” says McMullen of the Midwest ISO. “If the amount of wind that’s been financially cleared is significantly off from our forecast, we’ll use that information to make commitment decisions. We monitor that as the day goes on.”
Closer integration between various types of forecast data might someday lead toward a closed-loop approach to EMS that would require less hands-on decision making by control room operators. Such automation could provide quicker responses to fast-changing conditions on the system.
“The way power systems are operated today, they have quite a lot of human oversight,” says Lauby of NERC. “But the smart grid implies dispatching distributed resources in an automated way. Over time, with artificial intelligence and learning systems, it’s possible that dispatch signals on the bulk-power system will get somewhat more automated—under the oversight of skilled operators.”
Sullivan of KEMA agrees: “Europe already is heading in that direction,” he says. “In existing decision support systems, you could attach specific resources and demand to an algorithm. We have enough sophistication right now to close that loop.”
With increasing accuracy in wind forecasting, and increasingly intelligent grid and control-room systems, the concept seems genuinely possible. And as wind farms and other variable resources play a larger role, smart balancing of wind and load will become practical for both economic and reliability purposes.
“We’ve got a lot of wind on the system, and the lights stay on,” says Storck of 3Tier. “It hasn’t happened by accident.”
1. Eastern Wind Integration Study, prepared by EnerNex Corp. for the National Renewable Energy Laboratory, January 2010.
2. NERC’s Integration of Variable Generation Task Force (IVGTF) completed its draft report (NERC IVGTF Task 2.1 Report) on forecasting variable generation for operations and presented it during a NERC meeting in March 2010. At this writing, the NERC operating committee hadn’t yet approved the task force’s recommendations. Fortnightly reviewed the embargoed draft report for this article.
3. 130 FERC ¶ 61.053, Docket RM10-11-000, “Integration of Variable Energy Resources,” Issued Jan. 21, 2010.
4. See Central Wind Power Forecasting Programs in North America by Regional Transmission Organizations and Electric Utilities, Exeter Associates under contract to U.S. DOE/National Renewable Energy Laboratory, December 2010.