Solar and wind power plants provide clean, affordable electricity on a large scale, but they have one major scarecrow: intermittency. Essentially, intermittency refers to the fact that wind and solar power is both, to some extent, unpredictable and uncontrollable – and, in the context of a power grid, that means more controllable generating plants – like fossil fuel power plants – must be ready to fill gaps at all times so that customers do not experience any downtime. Now the UK – which depends on renewables for a plurality of its electricity supply – is turning to AI to reduce the functional intermittency of solar power.
In general, solar energy is less intermittent than wind, due to the predictability of day, night, solar intensity, and the angle of the sun to a location throughout the day. . Clouds, however, throw a wrench in the works, chaotically disrupting the solar power supply to solar panels with little warning (large-scale climate and weather forecasting models are, by and large, unable to solve the problem). individual cloud level). To complicate matters, energy operators, although aware of large-scale solar installations, are often unaware of the exact geographic location of solar panels on households or businesses. The combination of difficult-to-predict clouds and missing location information for many solar panels means operators don’t know when clouds are covering these solar panels – and, because of that uncertainty, the grid requires a larger buffer from other power sources to account for the difference.
Enter Open climate correction, a non-profit product lab “totally focused on reducing greenhouse gas emissions as quickly as possible”. The nonprofit has partnered with the UK’s National Grid Electricity Service Operator (ESO) to provide ‘nowcasting’ using machine learning. Essentially, the model will train a machine learning model to understand from satellite images how clouds move and where sunlight will fall. A separate project, meanwhile, is job to map solar panels in the UK.
âAccurate predictions for weather dependent generation like solar and wind are vital for us in operating a low carbon power system. The more confidence we have in our forecast, the less we will have to cover the uncertainty by keeping traditional and more controllable fossil fuel power plants, âsaid Carolina Tortora, head of innovation strategy and digital transformation at National Grid ESO. âWe are increasingly using machine learning to improve our control room forecasting, and this latest nowcasting project with Open Climate Fix – whose work could have real impact for network operators around the world – will bring another significant advance in our capacity and on our path to a zero carbon network.
This work builds on ESO’s machine learning efforts, which have already improved solar forecast accuracy by 33% in recent years.