Forecasting Solar Energy Using a Single Image

A team of researchers led by Jeremy Klotz and Shree K. Nayar from Columbia University has published a paper on arXiv detailing a novel method to forecast solar panel energy output using a single image. The technique, outlined in 'Forecasting Solar Energy Using a Single Image,' leverages computer vision to analyze visual cues from a photo taken at the panel's location. This allows the system to determine the camera's orientation and the portion of the sky visible to the panel, enabling accurate predictions of irradiance from direct sunlight and diffuse sky radiation. Additionally, the method accounts for reflections from nearby buildings, which vary smoothly over time and can be modeled from the image.

The approach was validated using real irradiance measurements in urban canyons, where it often outperformed conventional irradiance-based transposition methods and 3D model-based simulations. The researchers also demonstrated that a single spherical image can be used to determine the optimal fixed orientation for a panel. To facilitate practical deployment, they introduced Solaris, a device designed to capture the image seen by a panel in various urban settings. This work addresses the high soft costs of solar installation by eliminating the need for complex 3D modeling, potentially accelerating the adoption of solar energy in cities where rooftops and walls are increasingly used for panel placement.