Project start: 1.10.2025.

The large-scale integration of photovoltaic power plants (PV) into the power system requires accurate and rapid forecasting of their output, particularly over short time horizons. The FOTOVID project addresses this challenge by developing a system that utilizes wide-angle sky images and computer vision to forecast solar irradiance and PV output in the following minutes.

Current forecasting methods often rely on satellite data or numerical weather models with low spatial and temporal resolution, which limits their accuracy in local conditions. The proposed system is based on locally captured hemispherical sky images and measured solar irradiance, processed by a multimodal model using a CNN-LSTM architecture. This enables the extraction of visual features and modeling of temporal data sequences, resulting in more robust and responsive predictions.

The project includes the development and deployment of data collection modules at multiple regional locations to generate a new high-resolution dataset. These data will be used to train models for both local and regional forecasting. The system will be tested in real-world conditions, including validation at sites not involved in model training.

Expected outcomes include an open-access dataset, a prototype forecasting model, and scientific publications. In the long term, the project will lay the groundwork for collaboration with other institutions and industry, as well as future applications to competitive funding calls. Beyond its scientific contribution, the system has the potential to deliver meaningful societal and environmental impact—enhancing energy efficiency and supporting the integration of renewable energy sources into the power grid.