a cry for better forecasters in Denmark
When the wind blows over western Denmark at speeds between five and 12 meters a second (m/s), every 1 m/s change is equivalent to a power plant going on or off line. In other words, when transmission system operator (TSO) Eltra schedules power on the grid based on a wind forecast of 7 m/s, but the wind only blows at 6 m/s, Eltra needs to find 320 MW of power elsewhere to make up for the loss -- this amount is close to the largest single fossil fuel power plant in the system, the 400 MW Skærbæk facility. Additionally, every 1% increase or decrease in forecast accuracy can initially result in a gain or loss of DKK 2 million (g269,000) in balancing market costs.
"When meteorologists hear this, they fall out of their chairs," says Poul Mortensen, who is part of Eltra's team in System Operation, Analysis and Tools. "Their forecasting in the past has always been: will there be a storm tomorrow or not? Traditionally, they have made forecasts for air traffic, shipping and farming -- these industries are dependent on knowing how the weather will be tomorrow. But ours is a new market for meteorologists. Suddenly they have someone demanding quality in their forecasting, to get more precise wind speeds."
When scheduling power, Eltra uses a wind power forecast calculated before noon the day before the 24-hour operation period, which goes from midnight to midnight. Over a year, Eltra's average error rate in predicted production from wind plant, based on these forecasts, is 30-35% -- which means that the equivalent amount of power must be bought or sold on the real-time imbalance market. The TSO claims that the high miscalculation is not entirely its fault. "Calculating predicted wind speed into production is no problem," says Mortensen. "It's the easiest thing in the world. The hard thing is to get the meteorological forecast right. The quality of meteorological input to our systems is quite poor."
Eltra has one of the largest -- if not the largest -- percentage penetrations of wind power on a significantly sized grid in the world. In international research discussions of wind forecasting, Eltra is a major benchmark.
Eltra's grid, handling 60% of total Danish electricity consumption, covers the main peninsula of Jutland and the large island of Funen. It is relatively small, with maximum and minimum loads of 3800 MW and 1150 MW. Operating wind power capacity makes up about 2360 MW and occasionally the output matches total demand. Most of the capacity -- four out of every five megawatts installed -- is prioritised; market players are required to buy all production from these machines.
With so much wind integrated into Eltra's system, the value of wind forecasting is not even a question for discussion. "How do you put a value on a blackout?" asks Mortensen. "The production from this high amount of wind capacity has a great influence on the flow in our transmission grid."
Eltra began working with wind prediction models in 1996, when turbine installations were approaching 400 MW of combined capacity. That was when the TSO began to feel the wind on its system. Around this time, the Danish Technical University came out with its first version of the sophisticated Wind Power Prediction Tool (WPPT), one of two models Eltra uses daily. Now in version IV, WPPT crunches forecast numbers from several meteorological offices that deliver a new 48 hour forecast every six hours. This input is combined with a number of other factors as well as power measurements -- both historical and online estimates -- for statistical calibration. The other forecasting model in use, Frisk Vind (Fresh Wind), was developed by Eltra as a type of simple backup and cross-checking tool.
A day ahead
For every 100 MW of wind production, measured in quarter-hour intervals over a 24 hour period, Eltra's forecast from the day before deviates 30-35 MW on average, or 30-35%. "This varies a lot from year to year, depending on how stable the weather is. A year under 30% is good," says Mortensen.
From Eltra's figures in 2000, when wind energy produced 3384 GWh, forecasted miscalculations amounted to 1277 MWh, or 38%. Payment of real-time imbalance power that year was DKK 65 million, or DKK 0.02/kWh (g0.0027/MWh), which is the added cost for buying regulating power in the real-time imbalance power market. Mortensen says that wind power is responsible for most physical system imbalance in 70-80% of all hours.
He stresses that the 2000 figures are old and do not accurately represent the current situation. "The installed capacity has changed, the difference between the market and balancing prices has changed and forecast accuracy might have changed, too," he says, explaining that weather stability affects the results in forecast accuracy from year to year.
While a time-consuming analysis over more recent figures is not immediately planned, Mortensen has done a short analysis of 2003 data between January 1 and September 30 this year. In this period, of the 2350 MW of installed wind power capacity on the Eltra grid, an average 1878 MW was prioritised wind power that the utility was obliged take. From forecasts calculated before noon the day before production, the average miscalculation of prioritised wind energy predicted compared to actual production was 29%. Wind production for the period lies under normal level due to poor winds, he says, so the analysis cannot be used as a general indication of forecast error.
Misfires of 400-800 MW
When it comes to evaluating forecasting models in terms of money saved from balancing market costs, however, average values are not as important; here, what matters are the peaks and valleys of the actual errors, when Eltra has deviations of 500 MW to 1000 MW, says Mortensen. In about 90% of the first nine months of 2003, Eltra's forecasts were off by about 260 MW on average, with occasional misfires between 400-800 MW.
Denmark liberalised its power market July 1, 1999. Eltra acts not only as TSO but also as the market player responsible for production. Its own system of working the different markets is complex, but it can be boiled down to this. Three months before delivery, using past weather and production statistics as well as future supply and demand expectations, Eltra publishes a forecast of priority production. The TSO does this to make it easier for market players in Eltra's area to enter into future power purchase contracts, if they wish. Consumers are obligated to take this designated amount of priority power three months later.
The day before delivery, Eltra forecasts the next day's priority production using its wind power prediction tools. The difference between the three-month figure and the day-ahead one is traded on the Nordic spot market, NordPool, which closes at noon the day before. Up to and in the operation period, forecasts are continuously updated to help schedulers plan and balance supply and demand through trade in the real-time market. Settling the imbalance penalties among market players happens a week later on the balancing market.
According to Eltra figures from 2000, every 1% change in prediction accuracy results in a difference of DKK 2 million spent or saved on the balancing market. Applied to the initial 2003 figures, if the 29% error was reduced to 28%, that would mean approximately DKK 2 million saved in the balancing market. The savings will get bigger, however, the more the accuracy improves; the less regulation capacity is needed, the greater the savings. The more capacity needed for regulating imbalances, the higher the cost of electricity to the consumer
"At this point, it's very difficult to say if this figure has changed -- there are many factors to consider," says Mortensen. "We expect the regulation [real-time] market to develop over the years as well, so the span between the regulation price and the market price will be reduced.
Of the DKK 2 million for 1% deviation, he comments: "We do not often use this figure officially, because there is always someone who wants a piece of that cake." He continues: "We are interested in improving the safe running of the transmission system and market value for wind power by improving forecasts, but we no longer prefer to use traditional research institutes that live off funding rather than live off results. So we go out and pick the ones who we think can help us getting better forecasting results, and funding them is no problem."
One Eltra project for helping to improve predictions of output from wind plant is called ensemble forecasting. It seeks to provide more accurate and faster weather forecasts; the goal is to get more accurate wind power prediction. If the results come out as hoped, Eltra will be doing its own weather forecasting one day.
"Today, it's garbage in, garbage out. The most advanced wind prediction models in the world cannot change garbage into something valuable. It has taken too many resources and too much time over years instead of going to the real core of problem," says Mortensen. "The most important thing is to shake up the established meteorology society and make them understand that new requirements and accuracy are needed. Not only is it important to improve the accuracy of the meteorological forecasts but also to estimate the uncertainty of the forecasts. That can only be done with ensemble forecasting."