Photovoltaic (PV) systems are recently the most used sustainable energy source to fit with the energy demand growth. Generally, batteries, as storage systems, are installed along with PV modules. When it comes to an optimal power management of PV/battery hybrid systems, the uncertain and intermittent behavior of PV power production can provoke some challenges, with which, the real-time operation of the hybrid system can be degraded, therefore, PV power forecast is highly needed. Datadriven models are became nowadays very efficient methods to build regression models for the purpose of PV power forecast. In this paper, Adaptive Neuro-Fuzzy Inference (ANFIS) is chosen as a data-driven technique, to build up forecasting models. Standard ANFIS, which uses only weather data, cannot avoid the confusing scenarios like PV modules covered by the snow in clear-sky days. This work proposes an improved ANFIS model taking historical generated power into account. The developed model is validated on a real case, using the PV system of the institute of energy system technologies in Offenburg. When adding the average of produced power of the last 72 hours as additional input, the model was able to follow the rapid changes in weather conditions and overcome the unexceptional situations like the problem of snow on the PV modules.

8th World Conference on Photovoltaic Energy Conversion; 1253-1260