We're hiring in our ML team! 💻
Learn more
The transition to renewable energy is getting an extra boost from artificial intelligence (AI). Grid operators are starting to use innovative weather forecasting tools, based on AI, to help them better determine when the sun will shine on their solar panels and the wind will blow on their turbines - ensuring a steady supply of electricity. To drive this forward, Open Climate Fix has partnered with Google DeepMind. We are leveraging their AI-forecasting technology to further accelerate the energy transition.
To reduce the impact of climate change, humanity needs to dramatically and rapidly reduce our greenhouse gas emissions. Renewable energy sources like solar and wind are crucial, because they do not produce any greenhouse gases while in operation. Consequently, solar panels and wind turbines are being deployed at speed. The UK government has committed to decarbonise its electricity system by 2035, with the Clean Power 2030 Action Plan aiming to more than double UK solar power capacity by 2030.
The switch to these renewable sources of energy, and away from fossil fuels like coal and gas, brings challenges for the operators of electrical grids. In particular, solar and wind energy are inherently intermittent. Even passing clouds can dramatically reduce solar radiation and as a result, the output from solar panels. Likewise, the wind does not always blow - and sometimes it blows too hard, forcing wind turbines to shut down to avoid damage. To compensate for this shortfall, grid operators must make use of other sources of electricity. These challenges are becoming more acute as more renewable energy is installed, to the point of slowing down the integration of additional renewable sources.
Better weather forecasts, enabled by AI, can help grid operators manage this problem. The more warning an operator has of an upcoming shortfall due to changing weather, the more readily they can secure a backup electricity supply. By using these enhanced forecasts, grid operators can lower both carbon emissions and electricity costs, whilst also reducing the risk of blackouts.
The challenge is twofold: improve the weather forecasts themselves, and then translate them into predictions of renewable electricity output. Weather forecasts predict relevant variables like solar radiation and wind speeds, but they don’t forecast the amount of electricity generated from solar and wind. Open Climate Fix has addressed this problem by creating AI models that take input from weather forecasts and satellites. This enables us to generate high-quality electricity supply forecasts for grid operators.
Traditional weather forecasts operate solely from numerical weather prediction (NWP). They are based on equations that represent the physics of the atmosphere, which are then translated into algorithms that run on supercomputers. They represent a triumph of interdisciplinary work, enabling early warning of dangerous events like cyclones. Nevertheless, they have limits: notably, they struggle to predict peaks and troughs in solar electricity caused by shifting clouds.
However, a new generation of AI-based forecasts has emerged in recent years. These tools rely on machine learning: the ability of AI to learn patterns in enormous datasets. By training AI models on decades of weather data, it has proved possible to create highly accurate forecasting tools.
Google DeepMind released an AI weather forecasting model called WeatherNext Graph in 2023. It predicted weather conditions up to 10 days in advance and its accuracy surpassed that of the gold-standard High Resolution Forecast (HRES), produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). The following year, Google DeepMind released a new model, WeatherNext Gen. Again, it outperformed the best ensemble forecasting system of the ECMWF. It can forecast the weather up to 15 days ahead.
These new AI-based WeatherNext forecasting models require a fraction of the computing resources of numerical simulations, while achieving higher accuracies.
Open Climate Fix has focused on developing short-term solar and wind forecasts, which are crucial in supporting grid operators to quickly respond to changes in renewable energy generation. We created a machine learning model called PVNet that integrates satellite imagery and traditional NWPs, giving rapid forecasts of incoming solar energy. More recently, we built a second AI tool called Cloudcasting, an innovation that predicts satellite imagery of clouds.
Open Climate Fix is working with Google DeepMind to enable grid operators to use AI weather forecasts. Two projects are currently underway: one in the UK, one in India.
In the UK, we’ve been working with the National Grid Electricity System Operator (NESO) to improve their forecasting tools for the past five years. The UK, not known for clear blue skies, presents a challenge of accurately forecasting cloud cover. Our models have proven significantly more accurate at forecasting solar electricity production than NESO’s previous forecasts, increasing accuracy by 40% - further helping NESO towards its goal of a zero-carbon electricity grid.
We plan to leverage Google DeepMind’s technology to further enhance UK solar generation forecasting. While WeatherNext Graph and Gen outperform NWPs at predicting variables like windspeed, our work is identifying whether these new AI weather forecasts can also improve the accuracy of renewable electricity generation forecasts. If they can, our partnership has the potential to enable grid operators to save millions of dollars and tonnes of CO2.
Since 2024, we have also been working with a state grid operator in India on a similar project. India has an ambitious goal of installing 450 gigawatts of renewable energy capacity by 2030, posing a challenge for grid operators. We began by testing Google DeepMind’s forecasting models against regional historical data, and hope to implement the resulting forecast live in future. Our results so far have indicated a 10% reduction in large errors and a 5% reduction in our mean error, both across a forecasting horizon of 24-48 hours, which could translate into significant savings for the grid operator.
When scaled up, initiatives like these will help global grid operators transition to zero-carbon electricity quickly and efficiently, while ensuring a reliable supply and limiting costs.
Want to stay up to date with news from the project? Be sure to follow us on LinkedIn, or subscribe to our newsletter.
