Previously at the University of Cambridge, I have developped deep learning algorithms for advancing solar nowcasting based on cloud cover observations from sky cameras and weather satellites.
I'm interested in computer vision, machine learning, earth observations and solar energy. Much of my research is about improving solar forecasting tools from cloud cover observations. Representative papers are highlighted.
Forecasting solar energy accurately is good, generalising beyond the location where the model was trained is even better! Here, we explore diverse zero-shot and few-shot learning tasks based on a model pre-trained at another location.
We introduce SkyGPT, a physics-informed stochastic video prediction model that is able to generate multiple possible future images of the sky with diverse cloud motion patterns, by using past sky image sequences as input.
Literature review on the field of computer vision-based solar forecasting with a particular focus at the application of deep learning to process cloud cover observations collected by sky cameras or meteorological satellites.
Processing method for cloud cover videos to consistently centre the polar representation on the incoming flow of clouds using an optical flow algorithm.
Comparative study between various data augmentation and scene presentation methods including the use of polar coordinates. Application to solar forecasting from sky images and satellite observations.