Forecasting for Scheduled delivery

Forecasting wind strengths and predicting output from wind plant has

become a whole new wind industry business. But how much value does forecasting add to a wind kilowatt hour and in which circumstances does it add most? When does investment in forecasting pay off? And who stands to

benefit from it financially? In this ten page survey we seek answers in the markets with the highest penetrations of wind power

The more wind power installed on a power system, the greater the value of accurate forecasts of its output -- particularly when wind plant reach 10-20% of the system's total capacity. The financial benefits of that extra value should accrue to both the owner of the wind plant and the system operator -- and in a competitive market the savings should trickle down to mean cheaper electricity for the consumer.

For wind plant owners, the value lies in keeping costs down in so-called balancing markets, where players settle the difference for either producing too much or too little power than earlier promised. For Transmission System Operators (TSOs), the better the forecasts, the more accurately they can schedule all the generation needed, including wind power, and the less reserve power they need. Reserve power is used to smooth out all the fluctuations in supply and demand on the entire system. Good forecasts lower the costs of any reserve expressly added for backing up wind.

Wind forecasting has greater value where balancing markets are part of a competitive trading system for electricity than if forecasting is purely a tool to aid system operators with their balancing act. This is because the balancing market provides a financial incentive to both retailers and generators to get their output projections right -- and the more correct they are, the more efficient the system and the cheaper the resulting electricity.

The target

For system operators, a reasonable target for reducing costs through accurate forecasting is around $2.5/MWh -- with 10% wind on a network -- based on data from Britain's National Grid Transco and the US National Renewable Energy Laboratory (NREL). Indeed, evidence is mounting to support an estimate made years ago by the UK's Central Electricity Generating Board that if perfect wind power prediction was possible, it would add 10% to the value of wind power on the system. Today, such added value would allow wind costs in many circumstances to break-even with those of thermal power generation.

Forecasting also becomes increasingly valuable if system operators use long lead times in scheduling generating plants, as in Denmark (page 40). This is because it reduces the size of the likely discrepancy between projected output and actual output at the time of delivery.

Where forecasting does not appear to have great value is for the operators of large systems managing less than about 5% wind power. Deviations in wind output fail to show up in the ebb and flow of daily operation with small grid penetrations, even though the Spanish grid operator is suggesting otherwise (page 45).

The bottom line value of a prediction model is the reduction it is likely to achieve in the extra costs, either of additional reserve or balancing market penalties. Although there are variations between utilities, estimates for additional reserve are typically around $3-4/MWh with 10% of wind energy on a network, rising to around $5/MWh with 20% of wind energy. If these costs can be halved, as researchers claim, then the specific costs of adding an intermittent source of energy to the system such as wind come down accordingly. A halving of the extra costs for using additional reserve from, say, $5/MWh to $2.5/MWh narrows the gap between generation costs from wind and natural gas. Wind generation currently costs around $40-45/MWh on good sites, while gas-fired generation costs about $35/MWh.

fictitious costs

Balancing markets are an integral part of the new competitive electricity markets. If they are structured so that each and every deviation from scheduled delivery is apportioned to individual market players, they introduce fictitious costs (box), complicating assessments of the value of forecasting immensely. This is the case in Britain and in some areas of North America. Furthermore, the competitive markets tend to be based on bilateral contracts between generators and suppliers. The contracts are all different, making the advantages (or not) of improved predictability location specific. Electricity retailers and system operators tend to look unfavourably at technologies with variable and unpredictable output and may impose balancing market penalties, or require arbitrary discounts, such as those from the Bonneville Power Administration in the US (page 44).

UK consultants Garrad Hassan put it quite bluntly. The company had attempted to evaluate the financial benefits of forecasting for wind energy trading under the New Energy Trading Arrangements, or NETA, the market structure now operating in England and Wales, but this assessment, they stated, "cannot be done." The reason is that costs within balancing markets are volatile and critically dependent upon the exact rules governing particular markets. To make matters worse, the penalties that wind plant operators incur within balancing markets are also dependent on price differences between selling surplus generation and buying power to make up a deficit. These prices are also heavily influenced by trading strategies and the wholesale price.

Prediction methods

There are considerable difficulties in making accurate predictions of wind power output. Wind forecasting tools rely on numerical weather data input from meteorologists, whose forecasts are usually given over relatively large areas, mostly at ten metres height. Accurate forecasting requires input from a number of meteorological stations and sophisticated calculations to project forward to a specific location. For a given wind farm location, the wind forecasts then need to be turned into power forecasts, taking into account the topography of the site and the performance characteristics of the wind turbines.

For generators who trade wind in balancing markets, such as those in the UK and US northwest, most of the value of prediction comes from forecasts made in the few hours before production, since these markets close one to four hours prior to delivery. On the European continent, market liberalisation is not as advanced and instead of the system being controlled through the regulation of market forces, the shots are called by system operators. In Germany operators close the gates fully 24 hours ahead of delivery (page 42), and in Denmark and Spain they are respectively shut at noon and 10:00 before the next 24 hour period, which starts at midnight -- unfriendly action towards wind plant operators in both cases, which by its nature has a hard time predicting exact output considerably in advance of when it is needed. In the case of Denmark and Spain this is up to 36 and 38 hours ahead of time.

Western Denmark has a relatively small power system, but probably the highest wind penetration in the world, at times meeting all demand. It has to cope with fine tuning of supply and demand through the Nordic spot market, NordPool, which closes at noon the day before production. There, extra balancing costs depend on the difference between the spot market price and the balancing market prices (page 40).

As a result, system operator Eltra is likely to look upon total potential savings from wind forecasting more pessimistically than TSOs in less stringent market conditions. A tacit demonstration of this pessimism is offered by Eltra's Henning Parbo, who maintains: "It is not obvious that improved forecasting quality will reduce the necessary amount of regulating reserves and thereby reduce the fixed costs." He concedes, however, that "it is certain that improved forecasting of wind production will significantly reduce the activity in the balancing market and thereby reduce the energy price of regulation."


Much of the research into forecasting wind power concentrates on making better forecasts available on shorter time horizons, down to one hour. Wind variations are accounted for between the frames of "no prediction" -- where the system takes wind as it comes -- and "perfect prediction." No prediction is acceptable to system operators for low wind penetrations of up to 1-2%, since the variations from wind power will not be visible in the surges of supply and demand on the whole system. Perfect prediction is not achievable, but it is used as the ultimate benchmark.

Between these extremes is "persistence," which tends to be the default method of wind prediction, the measuring stick against which improvements in accuracy are assessed. In persistence, it is assumed that the power output from a wind plant at, say, 11:00 a.m. will be the same as it was at 10:00 a.m. Much of the today's research focuses on developing models that beat persistence.

According to NREL in America, requirements for back-up reserve capacity -- when the wind capacity amounts to 22.6% of peak demand -- might be reduced with accurate forecasts from 7.6% of the wind capacity to 2.6% (fig 1). This suggests that the costs of any extra balancing specifically for wind power can be reduced through forecasting by a factor of about three, with the exact monetary value depending on the costs of reserve. A realistic target for wind forecasts may be a halving of capacity requirements and costs. On a system with 10% wind, this translates to an operational penalty that comes down from around $4/MWh to around $2/MWh. The value of wind forecasting becomes more significant with high wind penetrations, rising to around $2.5/MWh with a power system's total installed capacity made up of about 20% wind.

There are two schools of thought as to where to place the costs of forecasting. Although the consumer ends up paying in the end, the responsibility depends on the structure of the market. In the US, it seems to be generally accepted that the responsibility lies firmly with the wind generators. That is also likely to be the case in Britain. In Denmark and Germany, however, the responsibility is shouldered by the system operator. There is some logic to this. System operators have for years used weather forecasts to aid their task of balancing the system, particularly forecasts of wind strengths since these influence demand for electricity for heating and other purposes.


Whoever channels the cost of forecasting through to the customer, there is no doubt that it has significant, though not huge, potential to improve the economic efficiency of systems fed with notable amounts of wind power. As Tom Osborne of the Bonneville Power Administration (BPA) in the American Northwest says: "Some have said that you cannot schedule the wind. While wind is not dispatchable, it is schedulable. By all accounts, scheduling accuracy of wind energy has improved over the past year at all of the wind plants in the Pacific Northwest. BPA closely monitors scheduling accuracy and performance.

"Others have said that you need the same amount of standby generation as the nameplate of the wind power plant. But this is not the case," Osborne adds. "By improving the accuracy of the day ahead and hour ahead schedule, power and transmission system operators are better able to maximise the flexibility, and value, of their systems," he concludes.

Have you registered with us yet?

Register now to enjoy more articles
and free email bulletins.

Sign up now
Already registered?
Sign in