Distributed solar prediction with wind velocity

The growing uptake of residential PV (photovoltaic) systems creates challenges for electricity grid management, owing to the fundamentally intermittent nature of PV production. This creates the need for PV forecasting based on a distributed network of sites, which has been an area of active research in the past few years. This paper describes a new statistical approach to PV forecasting, with two key contributions. First, we describe a 'local regularization' scheme, wherein the PV energy at a given site is only attempted to be predicted based on measurements on geographically nearby sites. Second, we describe a means of incorporating wind velocities into our prediction, which we term 'wind expansion', and show that this scheme is robust to errors in specification of the velocities. Both these extensions are shown to significantly improve the accuracy of PV prediction.