More wind power needs better forecasting

LONDON: One way to cut the cost of greater reliance on wind power is to improve day-ahead weather forecasting, to make it less expensive for grid operators to balance national demand and supply.
As countries seek to meet renewable energy targets, extra costs include subsidies, direct grid connection, backup reserves to cover intermittency and short-term grid balancing.
Short-term balancing adds costs both for grid operators, which have to pay power plants to turn off or to meet excess demand, and for generators switching on and off units designed for constant, baseload operation.
Renewable power will add to balancing costs by increasing uncertainty, for example when excess wind power has to be shut down on windy days or is dispatched to the grid unexpectedly.
Wind power in particular has an impact because its access to the grid is all but guaranteed, as a power source with zero marginal costs that can force out thermal fossil fuel generation.
One way to reduce that impact is to build out the transmission network to increase its geographic diversity, to capture a wider range of weather and so smooth intermittency.
The cost of widening the grid network is large, however, and it is time-consuming given lags in planning permits, fund-raising and cable tower construction.
A complementary approach is to improve forecasting skills, so that grid operators can more reliably balance the grid and less frequently call on more expensive, last-minute options including demand response (shutting down load such as supermarket cold stores) and gas and diesel generation.
Last year short-term balancing services cost Britain’s National Grid 708 million pounds ($ 1.12 billion).
The British think-tank, the Institute for Public Policy Research, last week published a report strongly in support of wind power, rejecting the idea that it was unpredictable.
The report, “Beyond the bluster, why wind power is an effective technology,” was in response to well-publicized skepticism among some parliamentarians in the Conservative party, the senior partner in the ruling Coalition, who have called for sharp cuts in subsidies.
Regarding the problem of intermittency, the IPPR report pointed out that baseload (constant) fossil fuel and nuclear power plants can trip instantaneously and without warning.
That illustrates an advantage of modular generation, where wind farms are made up of tens or hundreds of turbines, which are very unlikely to fail simultaneously.
On the issue of predictability, the IPPR report demonstrated how wind power does not vary significantly on the timescale of five-minute intervals, which is useful but less relevant than planning for weather variation hour by hour and the day ahead.
The day-ahead forecasting period is central to grid balancing because it corresponds with the start-up time for large, thermal coal-fired backup.
Regarding day-ahead variability, the study did not anticipate problems: “Wind power, at penetrations likely in the UK by 2020, is variable and predictable in much the same way as demand,” it said.
That confidence is not matched by Britain’s transmission operator, National Grid, which published a consultation earlier this year on whether to upgrade its wind power forecasting.
“The main challenge associated with wind power is its variability; wind power output is highly dependent on weather conditions and carries a high degree of uncertainty,” it said.
“As the volume of wind power capacity increases, so will the effect of wind variability and hence the accuracy of the wind power forecasts will become more important for both National Grid and the industry in terms of balancing their own position.”
The National Grid highlighted the problem of cut-out, for example, where high wind conditions force turbines to switch off, removing output suddenly: “These events are difficult to forecast accurately in terms of magnitude of impact and timing.”
The problem posed by wind power forecasting is not new, given that load is also weather-dependent, for example where the outside temperature drives heating and air conditioning demand.
While the IPPR report suggested that wind power was variable and predictable in much the same way as demand, a technical report by the US National Renewable Energy Laboratory (NREL), part of the Department of Energy, published in May found that wind power followed less regular patterns, including periods when there was no wind at all.
“This, of course, makes wind power more difficult to forecast than load,” the paper, titled “A Comparison of Wind Power and Load Forecasting Error Distributions,” found.
In the day-ahead time frame, the main cause of additional costs is the commitment or de-commitment of large thermal units, which means that big mistakes, while rare, are more important for system operations, given it is quite cheap to make small demand or supply tweaks.
For this reason, the thickness of tails — showing the proportion of extreme values — in the distribution of inaccurate forecasts is critical.
The NREL article found that wind forecasting errors by the Texas and California grid operators did not follow a normal distribution, instead displaying fatter tails and demonstrating that large, less frequent forecasting errors should be accounted for when calculating the system balancing costs of integrating more wind.
The paper illustrated the point with a one-week snapshot when wind output in Texas varied between 8,000 megawatts and almost zero.
It showed that as wind power output fell, the day-ahead forecast at one point over-estimated supply by 2,000 MW, a very large short-term balancing shortfall calling for more expensive fast-starting backup.
Similarly, Britain’s National Grid published forecasting shows that it underestimated wind power output by several hundred megawatts earlier on Monday, Sept. 3.
Given that patterns in wind power are not as predictable as demand, and have a shorter history of forecasts, it is clear that improvements in forecasting can cut the cost of system balancing.
— The author is a Reuters market analyst.
The views expressed are his own.