The Bonneville Power Administration (BPA) has developed a methodology that facilitates an hour-by-hour evaluation of wind forecasts, enabling grid operators to select the forecast that most accurately predicts wind profiles and provides the operators with a more precise tool to more effectively manage wind’s variability.
The methodology, called “Super Forecasting,” was developed by a small team at BPA to help the agency comply with two recommendations made by the U.S. Department of Energy (DOE). In 2011, DOE recommended that balancing authorities engage in a wind power forecasting program, and that balancing authority areas use more than one wind power forecast provider.
“Prior to December 2009, we had no wind power forecasting at BPA, and no situational awareness, so dispatchers on the transmission side and hydro duty schedulers had no idea where wind was going at any given time,” Scott Winner, project manager for BPA’s centralized wind power forecasting initiative, told TransmissionHub. “It’s terrifying to think that we existed like that, without wind power forecasting.”
In 2009, BPA began in-house forecasting, and subsequently contracted with two forecast vendors to align with DOE’s recommendations. However, forecasts from more sources don’t necessarily mean better forecasts. The question then becomes which one to use, Winner said.
In assessing results from the two vendors, Winner and his team sought answers to some basic questions: Should they use vendor A or vendor B? Are there times when it’s more appropriate to use one over the other? Or should the agency use A as a default, then go to B when A is unavailable?
The answer they reached was the methodology they have dubbed Super Forecasting.
“The Super Forecasting methodology is a neuronet, or an algorithm that we designed to blend A and B together so that we’re using both of them all of the time,” he said.
The algorithm performs an evaluation to determine which forecast vendor is offering the best performance by examining results from the past seven days at specific wind generation sites, then comparing the forecasts to the conditions that actually developed at those sites. Outcomes are then projected forward for the 72 hours of the next forecast period, resulting in a literal mosaic of forecasts for the facility.
“It may be that vendor A does better for the first six hours, then B does better for the next six hours, then revert to A,” Winner said.
The evaluation process is repeated every hour, going back over the past seven days’ worth of data to determine which vendor is providing the most accurate forecasts, and to determine what the new mosaic is going to be. The seven-day look-back period was chosen specifically to capture the most recent performance and evaluate which vendor is providing the most accurate information in the short term.
“We’re finding out that some vendors do better at certain [wind production] projects or in certain areas. Some are better at the short term; others are better in the long-term,” Winner said.
Rankings also change depending on the time of year.
“The weather that we have in August is very different than the weather that we have in May, and some vendors are better in August than they are in May,” he said. “That’s what the benefit of the program is: it allows us to tap into the best of what vendors are providing at that time.”
BPA has been running and testing the methodology for approximately a year and a half, though it was just placed into live service in March.
“In our program, we have three forecasts: two external vendors and an internal forecast generated within BPA,” Winner said. “BPA’s weather and stream flow wind power forecast is the baseline for performance. The expectation is that our vendors can do better than what we can do internally.”
Super Forecasting meets three objectives. The first is situational awareness for wind generation, and the second is load-resource balancing.
“Some of this wind actually serves load within our balancing authority area, so that generation needs to get modeled into our load resource modeling,” he said, noting that “We need to make sure the system is balanced with the wind that has been forecast.”
The third objective is to improve wind power scheduling.
The forecasts that are developed are transmitted to BPA’s dispatch center and the hydro duty scheduling center, enabling dispatchers to plan their transmission line loading based upon forecast information. They are also shared with BPA’s wind generators.
“Nobody is obligated to schedule to our forecast; it’s just an additional piece of information to help them with their scheduling,” Winner said.
Winner expects forecast accuracy to continue to improve as forecasters’ knowledge increases, a belief reflected in the request for offers (RFO) for new forecast vendors that will start in October 2014. That RFO closed May 20.
“We’re still figuring out wind power forecasting. Every time we do an RFO, we are updating the statement of work and asking for different things,” he said, noting that the RFOs have short contract cycles of two years to take advantage of rapid increases in forecast capability, as well as costs that are becoming increasingly competitive.
“The industry is maturing, as are forecasters’ processes,” he said.