Pv Production Forecast Using Hybrid Models of Time Series with Machine Learning Methods

Photovoltaic (PV) energy production in Western countries increases yearly. Its production can be carried out in a highly distributed manner, not being necessary to use large concentrations of so-lar panels. As a result of this situation, electricity production through PV has spread to homes and open-field plans. Production varies substantially depending on the panels' location and weather conditions. However, the integration of PV systems presents a challenge for both grid planning and operation. Furthermore, the predictability of rooftop-installed PV systems can play an essen-tial role in home energy management systems (HEMS) for optimising local self-consumption and integrating small PV systems in the low-voltage grid. In this article, we show a novel methodol-ogy used to predict the electrical energy production of a 48 kWp PV system located at the Campus Feuchtwangen, part of Hochschule Ansbach. This methodology involves hybrid time series tech-niques that include state space models supported by artificial intelligence tools to produce pre-dictions. The results show an accuracy of around 3% on nRMSE for the prediction, depending on the different system orientations.