With the widespread adoption of photovoltaic (PV) technology, distributed generation systems have become increasingly common, such as in residential areas. To run these systems efficiently, reliable solar irradiation forecasts are needed to help operators know what yields to expect on their solar panels – especially in the case of autonomous systems, when the solar panels alternate with a diesel generator.
Engineers and nature scientists at CSEM and Meteotest, working under a project sponsored by the Swiss Federal Office of Energy, developed a low-cost system that uses cameras and artificial intelligence to predict the amount of solar radiation expected at a given site in the near term. The project, called SkyCam, involved setting up CMOS cameras and high-precision pyranometers – solar radiation sensors – in Alpnach, Bern and Neuchâtel, three areas of the country with different climates and topographies. Data were collected every ten seconds for a year, along with solar radiation-measurements, painting a detailed picture of actual conditions out in the field.
An initial study in the framework of IEA PVPS Task 16 has shown that running the engineers’ data through AI algorithms generates relatively reliable short-term predictions of solar radiation, although the accuracy of the predictions still fluctuates. Now that the project is over, the team is opening up their unique, comprehensive dataset to the entire scientific community for use in developing short-term solar-radiation prediction systems based on data from local cameras. “We want to enable researchers to develop their own AI algorithms,” says Philippe Schmid, Head of R&BD, Industry 4.0 & Machine Learning at CSEM . “Switzerland was the perfect environment for us to build a vast dataset thanks to the country’s diverse topography and volatile weather conditions”, adds Jan Remund, Head of the Energy & Climatology business unit at Meteotest.
Head of Energy & Climate