Probabilistic Models for Photovoltaic Production Forecasting: A Case Study at the Politecnico di Torino Campus

Forecasting energy production from renewable sources is essential for the effective management of electricity transmission grids. However, the intrinsic dependence of these sources on meteorological conditions introduces significant uncertainty. If the characteristics of a photovoltaic (PV) plant are known, deterministic models allow production to be estimated with a reasonable degree of accuracy. When these characteristics are unknown, these techniques cannot be used.Assuming the availability of historical data relating to generated power and meteorological data measured in the vicinity of the plant (such as irradiance and temperature), energy production can be estimated using probabilistic models. This paper, therefore, presents an analysis of this specific case at the Politecnico di Torino Campus. Established and recently developed Recurrent Neural Networks (RNNs) were trained on consumptive data to evaluate their strengths and limitations. The results were then analyzed to determine future actions with forecasting data.