Forecasting focus on who pays
With more than 5500 MW of wind plant now online in Spain -- making up about 8% of total installed power plant capacity and around 5% of all Spanish generation -- wind forecasting has become a subject of intense interest and heated debate between the wind industry and the country's national grid operator, Red Eléctrica de España (REE). Fearing future problems with integrating ever larger amounts of an intermittent energy source, REE is demanding more and better forecasting of expected output from wind plant and money to pay for it. The wind industry is responding with a call for increases in wind power production subsidies if REE is expecting generators to carry the new forecasting costs (page 47).
Spanish regulations require REE to take all wind generation onto its grid, whether it is needed or not. REE schedules each 24 hours of production the day ahead at 10:00. To help it schedule wind output, REE uses a wind power prediction tool it developed called Sipreólico, which makes 48 hour and four hour forecasts for 80% of the installed capacity. The roughly estimated costs for maintaining back-up capacity reserves -- or "hot" reserves -- specifically for wind power are about i5/MWh, according to the national wind business association, Plataforma Eólica Empresarial. This expense is recouped on customer electricity bills.
"What is crystal clear is that we are not required by law to predict our production," says Alberto Ceña of the association. "We therefore need some kind of compensation to recover our expenses relative to the savings and benefits for the system."
REE's Cristina Martinez Vidal says the grid operator has already put great effort into improving wind integration. "A major problem is that wind production cannot be guaranteed," she says. Wind production can vary enormously after schedules have been drawn up. REE estimates the prediction error of Sipreólico's 48 hour forecasts at 20-30%. The error rate falls to 10% for forecasts four hours ahead of delivery.
While Sipreólico helps REE reduce power scheduling fluctuations across the entire national grid, continues Martinez Vidal, she expresses concern that reductions in local imbalances cannot be quantified. "Sipreólico could be used for wind to enter the [electricity pool] market if prediction was shifted to local level," according to Martinez Vidal. REE has begun work on a new version of its prediction tool with that in mind.
"Prediction is vital to the wind industry if it is to reach its 13,000 MW target for 2011," she adds. The unspoken concern is that if subsidies, or some form of structural incentive, are not available, generators will fail to make any effort to match projections of output with delivery. REE, it seems, may then resort to intermittent disconnection of wind plant in the name of grid security.
The legal implications are ambiguous. REE believes that supply security is the bottom line of 1997's electricity law. Wind generators believe the law's stipulation of renewables' free access to the grid to be paramount. Both sides hope that accurate prediction and money to pay for it will keep these two priorities compatible and away from the law courts.
Testing seven models
Meanwhile, seven wind project operators from Plataforma Eólica Empresarial have launched a i300,000 program to test seven different wind forecasting models over a year on 26 wind power stations averaging about 30 MW each. The program awaits financing, with hopes for a subsidy from the Spanish national energy program. The seven forecasting models each come from a different organisation. These include UK consultants Garrad Hassan, Danish state laboratory Risø, Germany's renewable energy research institute ISET, Cassandra Energy Services of Spanish Gamesa, Spain's public renewables institute Centro Nacional de Energías Renovables (CENER), Meteotemp of Switzerland and Meteologica of Spain, the last two international meteorological data providers.
"Our modelling effort is unique in the world because we concentrate on evaluating the impact of prediction on system costs and power guarantees," says Ceña. He says the program is also unusual in that it offers measurements right down to plant-by-plant level. Javier Sanz of CENER adds that the more complex the terrain, the greater the prediction error-and Spain has complex terrain. Why plant-by-plant testing is required on a fully integrated system -- which additionally does not penalise individual generators for deviations from scheduled production -- remains an unanswered question.
Once the test program better determines how far 24 hour wind prediction can be effective in terms of individual plant production and system operation, developers say they can calculate the size of subsidy or type of incentive they require to make output forecasts pay off. At the same time, the system operator and electricity distributors can calculate how much forecasting can save them in reduced levels of reserve power.
"It's early days yet to stipulate exact figures," says Ceña. The criterion for functionality is a 10% error rate for at least 50% of the operating time for each individual wind farm. The error margin will be largely reduced over a wider geographical area covering numerous wind plants. Some members of the working group think that the error factor will be achievable for 60% of operating time.
The seven companies offer highly varying prices for providing output forecasting for each of the wind plant in the program, ranging from i1000-i30,000 a year. The cost averages at i23,000/year for an average test bed wind plant of 28 MW. From here, Ceña estimates that forecasting would increase production costs by i8.6 million for the 5500 MW of wind plant operating today.
From its initial calculations Plataforma Eólica Empresarial is convinced that overall system savings through forecasting will exceed the costs -- though the actual amount of savings will only be clear after full testing of the various forecasting programs. At i0.60/MWh, forecasting over 2003 would cost i8.6 million, based on the group's estimated total wind production for the year of 14.4 TWh. Assuming that each kilowatt-hour deficit or surplus within the system adds i5/MWh for balancing power costs, forecasting would have to improve wind imbalances by an average 196 MW a day to cover its costs.
With about 5500 MW of wind currently online and an average capacity factor of nearly 30%, a minimum limit of 196 MW to make forecasting pay off is marginal and easily achievable, according to Ceña.