AI makes solar power predictable: New method improves yield forecasts

To efficiently integrate solar energy into existing power grids, it is crucial to accurately predict its output. A research team from Karlsruhe is working on improving this using modern AI technologies.

It is no longer sufficient to just know that photovoltaic systems generate electricity when the sun is shining. Grid operators, storage operators, and energy providers require precise information on when and how much electricity will be available.

Energy meteorology is dedicated to this task. The goal is to process weather data in such a way that reliable predictions about the solar energy input for individual facilities become possible. In addition to traditional weather models, algorithms from the field of machine learning are increasingly being used.

An example is a study conducted by the Karlsruhe Institute of Technology (KIT), which compared AI-supported methods with traditional statistical methods. Data from a Californian solar park over several years was used. The result: By refining the weather models using machine learning, predictions for solar power generation can be significantly improved, especially when optimization occurs at the end of the forecasting chain.

Particularly promising is a direct model that does not require detailed technical knowledge of the specific facility and can estimate energy production purely based on meteorological data. Considering the time of day in the learning process also significantly increased the accuracy of the predictions.

Although the study focused on a single facility, it provides important insights for the future handling of weather-dependent electricity production. With the further expansion of renewable energies, it will become increasingly important to proactively and reliably balance supply and demand.