Case Stories / Balancing energy grids using real time AI weather predictions

Balancing energy grids using real time AI weather predictions

Predicting weather is one of the most complex challenges facing humanity, and if solved would have some of the most far-reaching benefits. For businesses that play a fundamental role in everyday life, like agriculture and renewable energy, weather prediction is especially important. The difference between good and great forecasts amount to huge sums of money every year. With all that on the line, the latest artificial intelligence (AI) technology has become a beacon of hope to create a world in which perfect forecasts are the norm.

02/ In context

For the last 50 years or so, weather forecasts have been tackled the same way: extremely powerful computers crunch huge amounts of atmospheric and oceanic data. The companies that provide forecasts collect data from weather stations and combine it with data from multiple other sources like ocean buoys and independent weather trackers. The data is then processed using models that simulate the real physics associated with weather, which requires a huge amount of computational power, hours to be completed and a lot of money to collect and process this data. Forecasts are the most precise closest to the weather stations and exponentially become less accurate the further from the station you get. If your main concern is what jacket to wear outside, this may be accurate enough. But for many industries that impact the world’s economies, this is a serious problem.

03/ The challenge

Across the globe, farmers, air traffic controllers, financial planners and emergency agencies are just a few of the types of people that could seriously benefit from better forecasting.

For Swedish energy company Tekniska verken, imperfect weather forecasts are especially damaging, as they have to predict exactly how much electricity needs to be produced from their wind turbines and put into the energy grid 24 hours in advance. This prediction is what the system operator relies on when balancing the energy grid. When electricity grids aren’t properly balanced, this creates major costs for energy companies, often adding up to around 7 percent of the total profit.

“If the prediction is either too low or too high, we will suffer the economic consequences”, says Erik Olsson, business developer at Tekniska verken.

Unlike gas, electricity can’t be stored in large quantities, and thus the ordered demand and supply must balance every day. With wind energy, both the demand and the supply – the wind itself – can be better predicted with stronger weather forecasting, thanks to AI.

This predicament extends into electricity trading as well. Olsson explains, “If you know exactly how much the wind turbines will produce for, let’s say, the next 48 hours, you can sell it on the market for a higher price. This makes extremely detailed forecasts so valuable for us.”

Tekniska verken is using Peltarion’s cloud-based AI platform to develop a deep learning model to improve weather forecasting. They’re calling it Deep Weather as it identifies patterns in both live and historical weather data using cutting-edge machine learning algorithms. Deep Weather’s ability to process larger amounts of data with better and faster computing can help Tekniska verken steer into a new era of weather forecasting.

04/ The results

Deep Weather’s innovative method of weather prediction has already proven to be both faster and cheaper in comparison to the traditional approach, which takes three hours and requires €30 million in computer hardware to process. Deep Weather can make the same prediction in 100 milliseconds on a computer that costs €10,000, which is a massive improvement.

For a wind park owner, this is great news. Tekniska verken’s operators can send weather data from their windmills to a meteorological institute, and receive a customized forecast for their exact location, specifying exactly how much energy needs to be produced. This means that accuracy can be greatly improved, all done in a matter of minutes.

For Tekniska verken, error margins can potentially be reduced from 14 percent down to 5 with the use of Deep Weather. Olsson knows this is huge for their prospects: “Of course, this could save us a lot of money, but also, it makes investments in wind power far more attractive. And the model only gets better and better, when more data is fed to it”.

05/ Conclusion

Deep Weather was created with wind energy optimization in mind, but the system has the potential to create all sorts of new solutions, large and small. Office buildings could be provided with updated forecasts every 10 seconds and automatically adjust their climate systems accordingly saving billions in energy waste. Off-piste skiers and mountain climbers could enter the wilderness with full confidence that adverse weather won’t leave them stranded. For Olsson and Tekniska verken, the quest to be more resource efficient is a never-ending one. “We have an obligation to always challenge the boundaries of what is currently possible. Deep Weather has demonstrated that it is possible to rethink how to solve complex problems with AI, and that it’s possible to leverage the downsides facing wind power into advantages that will ultimately lead to more renewable energy”.

Photography by Paulo Simoes Mendes, Drew Hays and Samuel Wester