A new innovation project will examine how AI can be used to improve how the grid forecasts solar generation.
National Grid Electricity System Operator (ESO) is working with non-profit start-up Open Climate Fix (OCF) to develop a first-of-its-kind solar ‘nowcasting’ service for its national control room. This involves using a machine learning model that can forecast the near future, giving forecasts in minutes and hours rather than days.
Nowcasting has historically been used for rainfall prediction, but OCF will take a similar approach to predicting sunlight. The company is training a machine learning model to read satellite images, and understand how and where clouds are moving in relation to the solar arrays below.
Previously there has been no way to anticipate short term swings in solar generation due to cloud cover, and this uncertainty is part of the reason the ESO keeps reserve power sources on the grid. This often comes from flexible gas plants, which are both heavy polluters and expensive.
By having more accurate forecasts, National Grid ESO could reduce the number of these plants it relies on, instead using more efficient balancing actions that would be better for consumers and aid the shift to a net zero system.
“Accurate forecasts for weather-dependent generation like solar and wind are vital for us in operating a low carbon electricity system,” said Carolina Tortora, head of innovation strategy and digital transformation at National Grid ESO. “The more confidence we have in our forecasts, the less we’ll have to cover for uncertainty by keeping traditional, more controllable fossil fuel plants ticking over.”
The partnership with OCF forms one of a number of projects National Grid ESO is undertaking to boost its forecasting abilities. For solar specifically, this is being aided by work to map out ‘invisible’ solar panels, which it is undertaking together with the University of Sheffield’s PV_Live initiative.
Earlier this month, the operator announced it is exploring the possibility of moving to dynamic day-ahead reserve setting as part of a new project with Smith Institute. This will use a machine learning model based on predictor variables like temperature and wind forecast data that can create more accurate predictions of forecast errors, accounting for the variability in weather on a day-to-day basis.
This work all builds on the ESO’s previous work using a ‘random forest’ model to improve solar forecasting, which a report in 2019 from the Alan Turing Institute showed had its solar forecasting accuracy improved by 33%.
“We’re increasingly using machine learning to boost our control room’s forecasts, and this latest nowcasting project with Open Climate Fix – whose work could have real impact for grid operators around the world – will bring another significant step forward in our capability and on our path to a zero carbon grid,” added Tortora.
OCF was co-founded by former DeepMind researcher Jack Kelly, and in April was awarded £500,000 for its solar power forecast technology from Google.org. This grant has allowed the company to further adapt and apply elements of its “transformer” models to the solar sector, having previously used it in predicting the shape of proteins.
“We're over the moon to be collaborating with one of the world's most innovative system operators – National Grid ESO,” said Kelly. “We plan to adapt the amazing work done by the global machine learning community to solar electricity forecasting. All our work will be open-source, so others will be free to use the technology to help reduce emissions globally as rapidly as possible.”
Solar Power Portal recently caught up with Kelly to discuss exactly how this transformer model works, and what it could mean for the solar sector in the UK.