A new wind-power forecasting system could help National Grid cut the costs of bringing more turbines into the electricity network.
The system uses more complex computer models to predict power output from windfarms with greater accuracy and also gives a better picture of how accurate the forecasts actually are.
This will allow National Grid to be better prepared for times when the wind doesn’t produce as much electricity as predicted by having the right levels of backup generation to balance supply with demand.
‘We have to procure those services from the market and spend money to do that. And ultimately it’s consumers who bear that in their bills,’ said Richard Smith, National Grid’s future transmission networks manager.
‘Being more accurate on the wind-power forecast allows us to better manage what we’re doing. If we can better manage the actions we’re taking, it allows us to minimise the number of balancing actions and therefore reduce the overall cost.’
Wind power is expected to go from less than 5GW of generating capacity to around 28GW by 2020 — representing one quarter of the UK’s energy mix as coal plants are taken offline.
The intermittent nature of wind turbines will inevitably increase the cost of balancing supply as more windfarms are built, as will plans to change demand patterns, for example, by installing smart metersand using more off-peak energy.
National Grid expects annual balancing costs to rise from around £280m to £500m by 2020, even with the new forecasting system. But this means that even relatively small improvements in accuracy could save millions of pounds.
In its first two months of operation, the new system has produced power-output forecasts that were on average 14 per cent too high or low. The old system running parallel had an average error rate of 18 per cent.
The new system works by taking four wind-speed forecasts at regular intervals throughout the day and running them through three computer models. These 12 forecasts for each location compare with the one currently produced.
The computer program also produces forecasts for a higher number of locations and can predict the wind speed for regional zones of windfarms instead of just for individual sites or the UK as a whole.
National Grid’s existing system uses a standard physical model that relies on a straightforward equation to calculate power output from wind speed.
The new program also uses a statistical model based on a year’s worth of recorded data on wind speeds and an artificial neural network that can learn over time to improve its calculations. One day it may also include data from individual turbines.
‘The only drawback is that it learns everything you teach it,’ said senior forecasting analyst David Lenaghan. ‘If you have any bad data — sometimes our metering systems drop out — then the system learns it.
‘This system gives us a lot more avenues to be more accurate, but we’ll need a lot more work to achieve that as well.’
The new weather forecasts used also predict a range of wind speeds, rather than a single average. This provides the system with a better idea of how accurate the power-output forecast will be, because smaller ranges leave less room for error.