The role of generative AI in the energy transition

Solar, wind and hydropower are key drivers in the energy transition. Electricity production from solar and wind energy is subject to significant fluctuations. These fluctuations must be mitigated through the use of storage systems and conventional power plants such as gas-fired, nuclear and coal power plants, whose output can be adjusted to meet grid demands. To ensure grid stability, the amount of electricity fed in must always correspond to the electricity consumption. And to feed significant proportions of renewable energy into the grid, accurate forecasts of expected electricity generation from solar and wind energy are necessary. Researchers at Bern University of Applied Sciences are working on improving the predictability of renewable energies in order to maximise their use and thus make a contribution to the energy transition.

New models that revolutionise weather forecasting

“Artificial intelligence and real-time satellite data are the key to a better integration of solar energy into the grid”, stresses Angela Meyer, Head of the AI for Energy and Industry Group. The new forecasting models that are developed worldwide represent a real revolution in renewable energy generation forecasting. In a project funded by the Swiss National Science Foundation SNSF, Angela Meyer’s research group has developed a model that predicts solar irradiance and extends the intraday solar forecast horizon for power generation by one hour. The short-term generation forecasting has been extended by one hour without any loss in accuracy. SHADECast, a model for solar irradiance forecasting that was developed for this purpose, relies on near real-time satellite data. Integrating machine learning, the model makes cloudiness and solar irradiance forecasts with lead times ranging from minutes to hours. Trained with 10 years of satellite data, the model can make precise short-term forecasts of solar power generation. Until now, many solar power forecasts have relied on classic weather forecast models. However, these are often less suitable for precise short-term forecasts, as they tend to lead to large errors in intraday forecasts, especially in changing or cloudy weather conditions. More precise forecasting allows producers to optimise their production and electricity trading strategies, while grid operators can ensure solar power feed-in and grid stability by planning the use of balancing energy more precisely and by preventing fluctuations in the grid.

Actively shaping the energy transition

Research carried out by the AI for Energy and Industry Group at the Institute for Data Applications and Security IDAS encompasses yield forecasting and data-driven asset management for large energy parks powered by renewable energies, including improved operation, maintenance and predictive maintenance, as well as increasing the profitability of a system. In addition to the solar forecasting models, a new project for the short-term forecasting of wind energy has recently been launched with the support of the SNSF, and further projects are in the pipeline. Angela Meyer expressed her satisfaction with the level of interest shown. “It is very encouraging to see that our work is actively contributing to the energy transition, and that interest in it is growing.”

AI can also contribute to long-term forecasting

Long-term forecasts are also required. Investors in energy parks are interested in what production conditions will look like in the future. For example, they want to know how much sun and wind there will be on a given site in the future, whether the installations will have to be built more robustly because more severe storms are expected, and whether periods without sun and wind will become more or less frequent due to climate change. “With our research, we can also make a contribution in this area,” says Angela Meyer.