Comparative study among physical models and artificial intelligence methods for PV power forecasting

The renewable energy is widely used around the word, thanks to its multiples benefits as property and durability. Upgrading the accuracy of photovoltaic (PV) power forecasting is critical for management and dispatching operations. Nevertheless, this is a difficult task due to its intermittent nature. As the photovoltaic power depends mostly on solar irradiation and ambient temperature, its prediction can be done through meteorological variables estimation. Various methodologies have been developed, including physical models and artificial intelligence techniques. This paper provides a comparison between empirical models (Perrin de Brichambaut, Liu and Jordan, Eufrat, Ghouard, Capderou and the ambient temperature distribution model) and the deep learning techniques (the adaptive neuro-fuzzy inference system (ANFIS) and the multilayer perceptron neural network (MLP)) to explore their effectiveness in PV power forecasting based on the errors percentages. The simulation results show that the artificial intelligence topologies outperform the physical models. Moreover, the ANFIS model is more accurate than the MLP network.