The Canadian province of Alberta is proving to be a tough testing ground for the accuracy of wind power production forecasts. The region's complex topography and the impact that has on weather patterns is stretching the abilities of both European and American forecasting experts to their current limits. Just how much accuracy can be achieved in Alberta still lies in the realm of the unknown, but the results of a year-long study have convinced both the industry and the Alberta Electric System Operator (AESO) that forecasting is essential to integrating increasing amounts of wind on the province's power system.
"I think it is a key and a fundamental piece. I think any jurisdiction that is integrating wind to any extent -- and that's everybody in North America -- needs sophisticated forecasting tools, and those tools need to be integrated into the day-to-day operations and the decision support tools the operators use," says Warren Frost, AESO's vice president of operations and reliability. "From an AESO perspective, what's important for us is that wind forecasting becomes part of our base business, that it is woven right into our normal operating practices."
The study, designed to test various forecasting methods and determine the most effective one for Alberta, started in May 2007 and ran until May this year. Three companies provided forecasts for seven existing and five future wind power facilities located in four different regions of the Canadian province. The forecasts covered the next 48 hours and were refreshed every hour.
Mississauga-based Ortech Power was given the job of providing an independent analysis of their performance. At a final workshop in June that was broadcast over the internet to about 100 listeners around the world, the company's Don McKay estimated that his team had to sift through about 100 million points of data. The bottom line, he said, is that it is possible to forecast wind in all four specified regions. "Will it be easy? No. It's going to be difficult." All three vendors, who employ very different methods, came within the range of forecast error found in the literature or "maybe a little larger because of the complexity," said McKay.
The analysis also shows that forecast accuracy decreases moving out in time from one to 48 hours and that a persistence forecast, which assumes whatever happened in one hour would happen for the next 48, tended to be comparable to the forecasting models up to six hours prior to delivery. Forecasts were least accurate in the afternoon between 13:00 and 18:00 and in the winter months of November through February. The most accurate forecasts of wind speed were made in the central part of the province, while the least accurate were recorded in the southwest, the region that is closest to the Rocky Mountains -- where most of Alberta's wind facilities are currently located.
The ramp not the ripple
None of the forecasters, said McKay at the workshop, had much success capturing either the timing or magnitude of extreme up or down ramps in wind power production. The findings around ramps, which can pose significant system operational challenges, were particularly important to AESO.
"As you go forward and get more wind you get more variability on the minute-to-minute basis, but that's not the issue. That is not going to be an integration issue. I think that is a learning issue for not only us in Alberta but across North America, realising it's the ramp, not the ripple," says Frost. "What that starts to tell you is, if that's your key operational issue, then your forecasting needs to be either tuned towards it or present results in a way that ensures the operator understands that he may have a ramp and that there is some uncertainly around it. Knowing the uncertainty is good too."
For the forecast companies involved, the Alberta challenge was somewhat unexpected. "It was more work than we thought at the beginning. It was much more work," Ulrich Focken of Germany's Energy & Meteo Systems told the final workshop. Corinna Mohrlen of Danish company Weprog agreed. "We didn't realise how really difficult it is to forecast in Alberta." John Zack of New York-based AWS Truewind told the workshop that the "frequent occurrence of difficult wind regimes to forecast" is one reason forecasts have a mean absolute error 20-30% higher in Alberta than other jurisdictions in North America.
But at the same time, all three vendors believe improvement is possible. AWS Truewind, for example, experimented with adjustments to its system to account for shallow cold air events, a common phenomenon along the Rockies, and got a substantial improvement in accuracy. "This is an example of what you learn and the impact it can have on forecast performance," said Zack. Weprog tested higher resolution models in the final three months of the study and found significant improvement in forecasts. "The good news is we do understand now what is required. We do understand and know where we have to improve," Mohrlen said.
Weprog also did some analysis of wind ramping events and found no systematic patterns. "That means as a consequence it will be very subjective to correct any phase error. What that means is the operator in Alberta cannot expect sustained high accuracy forecasts for the day ahead." Focken, though, believes a better understanding of weather patterns responsible for the ramps, and adjustments in the models to reflect that, will help.
"There is improvement in ramp forecasting possible if the requirement is to ramp forecast. But that was not the requirement at the beginning." Numerical weather predictions provided by organisations such as Environment Canada or the US National Weather Service, which all forecasters work from, actually provide a lot of information about ramp events, added Zack. But what forecasters need, he says, are the tools to pull it out and make it useful.
One of the problems with counting on numerical weather predictions is that they are not really designed to provide the specialised data that wind forecasters require. Environment Canada, said the department's Serge Besner, is trying to tackle that issue by analysing how it can better serve the wind energy sector.
"There seems to be a shift in the meteorological sector in listening and finding out what the needs are and, perhaps, catering our products for the different sectors," Besner said. "We won't have the magic bullet for everybody. But certainly we're willing to listen to what is required from Environment Canada, in terms of products and services as well as our data -- and with the resources we have available make the changes that can benefit the energy sector."
Learning a lot
In Quebec, Environment Canada is already working on a project to develop a high-resolution weather prediction model that can be used for wind power forecasting in that province. In a final report issued last month, the industry work group that steered the Alberta study recommends that wind facility owners and AESO work with Environment Canada to help validate the model. Zack would also like to see a continuation of the Alberta-specific research that started with the forecasting study. "All of the forecast providers have learned a lot and there is a lot more to be learned," he said.
One of the study's funding parties, the Alberta Energy Research Institute (AERI), is open to the idea. "When AERI got involved with this project, we hoped the outcome would be the beginning of a forecasting system that would allow the AESO to call on wind power on a reliable basis. What we got though, not to minimise the great work that was done, is a reminder of the unpredictability of the wind and especially the complexity of wind patterns when you live next to the mountains," says AERI's Richard Nelson. "But this is the start, we think. We at least have recommendations on the direction we have to go on further research to achieve the ends that I talked about," he adds.
What is also important, says the work group's final report, is getting a forecasting service in place as soon as possible to give the AESO an opportunity "to learn how to better use a forecast for system operations." One of the keys to that, it adds, will be to first determine how the forecast will be used. "The pilot project demonstrated that without this focus, the nature of forecast error may be too broad for one single forecast to be optimal for multiple purposes such as real time operations, transmission scheduling and ancillary service forecasting."
With the study complete and the work group recommendations in hand, says Frost, the next step is for AESO to develop a discussion paper on how it will proceed and then consult with market participants. The grid operator wants to have a forecast system in place by the time wind capacity in Alberta gets to 700-800 MW, says Frost. "Current forecasts are saying that we'll be there at the end of 2009, early 2010. It is hard to say because it is very much dependent on the development schedules of developers and on transmission as well."