Elia Group helps businesses and consumers reduce carbon emissions with Vertex AI
Oz Ural
Product Lead, Elia Group
Editors note: Today we’re hearing from Oz Ural, Product Lead at Elia Group, about the Group’s innovative approach to measuring and forecasting the CO2 intensity of electricity across their grid. This is helping businesses make informed decisions about their energy consumption.
The energy sector is facing huge, exciting challenges. The transition from fossil fuels to a sustainable energy supply is a complex process, which requires the involvement of many different actors. Indeed, given that more than a third of the world’s electricity is due to come from renewables by 2025, the ongoing shift to a more sustainable future requires collaboration, innovation, and determination. As part of this shift, we at Elia Group want to establish the electricity grid of the future using state-of-the-art technologies.
Through our subsidiaries in Belgium (Elia Transmission Belgium), and the north and east of Germany (50Hertz), we operate 19,460.5 km of high-voltage connections, meaning that we are one of Europe’s top 5 transmission system operators. Key projects of ours are focused on the expansion and reinforcement of our grid and the integration of increasing amounts of on- and offshore renewable energy into the system. But that’s just for starters.
Tackling carbon emissions by measuring CO₂ intensity
As we work towards a more sustainable future, we want to give everyone access to data that can help them make more informed decisions about how to reduce their carbon dioxide (CO₂) emissions. We’re doing this through measuring what’s called ‘CO₂ intensity’. This is used to describe the amount of carbon dioxide that is emitted per unit of energy generated, and is expressed in grammes of CO₂ equivalent per kilowatt-hour (gCO2eq/kWh). It is often used to measure the environmental impact of energy sources, such as electricity, meaning it is a measure of how ‘clean’ electricity is.
For any given amount of electricity, the lower its CO₂ intensity is, the less carbon dioxide it emits. Renewable energy sources have a much lower CO₂ intensity than fossil fuels. This means that, during moments when it is not windy or sunny, fossil fuel power plants (for example, those which use lignite or gas) have to generate more electricity, causing an increase in carbon emissions. On the other hand, if it’s a sunny or windy day, CO₂ emissions are relatively low or nearly zero. This is only currently the case in countries which have a lot of wind and solar farms, like Germany. However, as the use of renewable energy increases across Europe, more and more countries will need to measure the CO₂ intensity of their electricity.
What does this mean in practice? Let’s say it’s windy in Berlin, but it’s also really overcast. It’s likely that the CO₂ intensity of the electricity that the city is using is quite high, because fossil fuel power plants have to work harder to make up for the lack of solar power. However, if the weather forecast predicts that there will be a big storm in the evening, it’s likely that the CO₂ intensity will be much lower later that day, since there will be a surge in wind energy which will mean the city will be able to rely less on fossil fuel power plants for its energy supply. Berlin’s citizens will then know that they should charge their electric vehicles in the evening, so saving on carbon emissions. This detailed information about the CO₂ intensity of electricity across the grid can help companies and their customers improve their understanding of their CO₂ footprints and find ways to reduce their CO₂ emissions.
Forecasting CO₂ intensity with Vertex AI
Elia Group is committed to the idea that innovation is key for a successful energy transition. In line with this, we have our own ‘incubator’ called The Nest. Our work at The Nest is helping develop and create new solutions that will help to boost Europe’s establishment of a sustainable energy sector. To do this, we’ve been working on a way to measure the CO₂ intensity of electricity across our grid. We call it eCO₂grid.
Initially, our forecasts in this area were relatively limited in terms of functionality and geographical scope. We therefore wanted to expand our forecasting tools to encompass a broader range of data. Having already tested out Google Cloud’s Vertex AI as part of several other projects, we began to think about how it could help us construct eCO₂grid. The Google Cloud team connected us with Google Cloud Partner Eraneos to help build eCO₂grid. We soon realised that Vertex AI was the solution we needed to help us improve our CO₂ intensity forecasting as part of eCO₂grid.
Building an ML Ops platform at scale
Along with Eraneos, our data scientists and engineers built a comprehensive MLOps platform using Vertex AI which incorporated all the existing forecasting models and connected them to the data in our systems. This allows us to control and manage multiple models at the same time, which wasn’t an option that had been available to us previously. This has enabled us to develop advanced 24-hour forecasting models for our home markets of Germany and Belgium, with the possibility of extending these models to include each of Europe’s electricity market areas (which are called ‘bidding zones’ and are mostly aligned with national borders). This is very important because each bidding zone has such a diverse energy mix which cannot be easily incorporated into one single model. That's why being able to handle multiple models at the same time is so useful, since this allows us to produce much more accurate CO₂ intensity forecasts.
Building this platform using Google Cloud was quick. We put together an initial model with our data scientists in a few weeks. By the end of 2023, we had a model that was as accurate as we could have imagined. We’d like to receive more feedback about how useful eCO₂grid is for our customers and we hope that, eventually, our forecasts will become the gold standard for Europe. We want companies who are producing sustainable products, whether that’s electric vehicles or heat pumps, to use our forecast data in their own systems via an API. Ultimately, we want to give everybody a better understanding of what kind of energy is produced where and what impact it has on the environment.