Knowing the photovoltaic (PV) energy produced at a given time and a specific position is crucial to handle renewable energy. Nowadays, Machine Learning (ML) is broadly used to predict energy production and consumption, but existing datasets cover only some regions in each country, which impedes the deployment of such systems in production. This paper proposes a novel generalized geolocalized methodology to predict photovoltaic production for the next hour and the next day at a given geographical point. This result is part of the deliveries of a Swiss Federal office of Energy (SFOE) project called Micro Storage Intelligent and Distributed (MSID).