Britain’s dreary weather is the subject of many jokes, but the UK’s National Grid is making the most of the sunshine.

By using Open the Climate Fix appnon-commercial product laboratory, control room National Electricity Network Operator (ESO) tests artificial intelligence models that provide accurate short-term forecasts of sunny and cloudy conditions over the country’s solar panels.

The information could help ESO, the UK’s electricity grid operator, tackle a key problem in renewable energy: sudden cloud cover can cause a significant drop in solar output, prompting grid operators to ask fossil-fueled plants to recycle power as a backup.

With better forecasts, ESO can cut additional fossil fuel energy stored as a reserve, increasing efficiency while reducing its carbon footprint.

“Traditional weather models don’t predict clouds very well, but by using artificial intelligence and satellite imagery, we can bring much more precision to the solar forecast,” said Dan Travers, co-founder of Open Climate Fix, a UK-based startup. “Solar is really effectively displacing coal, but grid operators need accurate forecasts to integrate large amounts of solar generation, so we see a lot of opportunity in applying this solution to coal-fired power grids around the world.”

Open Climate Fix is ​​a member The beginning of NVIDIA, a global program that offers the latest startup experience, technology and go-to-market support. The team publishes its datasets, dozens of models, and open source code HuggingFace and GitHub.


Each colored dot on the map represents a solar PV system. Blue dots indicate low solar output, yellow dots indicate high output, and black dots indicate systems with no data.

AI to catch the cloud and pin it

Before the advent of renewables, the experts who managed the electricity grid on a daily basis only had to worry about the variability of demand on the grid—making sure that enough electricity was being produced to keep up with air conditioners in the heat of the day or electric stoves and appliances on weekday evenings.

When adding renewable energy sources such as wind and solar power, the power grid must also take into account changes in supply levels due to weather. Satellite images provide the most up-to-date picture to determine when clouds come between the PV panels and the sun.

Open Climate Fix’s AI models are trained on terabytes of satellite data acquired at five-minute intervals over Europe, the Middle East and North Africa. Additional data sources include hourly weather forecasts for years at 10km resolution, topographic maps, time of day and sun position in the sky, and live readings around solar panels across the UK.

The team uses some of the latest deep learning models for weather modeling, including MetNet, GraphCast and the Deep Generative Model of Radar. They showed that their based on the transformer The artificial intelligence models are three times better at predicting solar energy production than forecasts produced by ESO’s traditional methods. The increased accuracy could help ESO achieve its goal of being able to control a zero-carbon electricity grid by 2025.

“Physically-based forecasting models are powerful for predicting weather on the scale of days and weeks, but take hours to build, making them unsuitable for hourly or minute-scale forecasts,” Travers said. “But with satellite images taken every few minutes, we can get closer to a direct image of cloud cover.”

AI works on Sunshine

Cloudiness is of particular concern in the UK, where cities including London, Birmingham and Glasgow get an average of 1,400 or less hours of sunlight each year – less than half Los Angeles. But even in desert climates where cloudy days are rare, Open Climate Fix’s AI models can be repurposed to detect when solar panels are covered in dust from a sandstorm.

In addition to forecasting for the whole of the UK, the not-for-profit organization is also developing models that can predict how much energy individual solar panels will produce. This data can help operators of large solar farms understand and maximize energy output. Smart home companies can also use this information to optimize energy use from customers’ rooftop solar panels, giving homeowners information about when to run energy-intensive appliances or schedule electric vehicle charging.

Open Climate Fix uses a cluster NVIDIA RTX A6000 graphics processors provided via an NVIDIA hardware grant feed his work. When training multiple models simultaneously, the team shifts its excessive workload to NVIDIA A100 Tensor Core GPUs available through cloud service providers.

“The hardware grants helped us develop and iterate our models more easily,” said Jacob Beeker, a machine learning researcher at Open Climate Fix. “When our team first debugs and trains a model, it’s twice or three times faster to do it locally.”

To learn more about AI accelerates decarbonizationimproving network resilience and improving energy efficiency, register free for NVIDIA GTCwhich takes place online on March 20-23.

Read about NVIDIA’s work in energy and utilities and apply for membership The beginning of NVIDIA.

Main image National Grid ESO Electricity National Control Centercourtesy of the ESO Media Center