Modeling of photovoltaic modules using a gray-box neural network approach

This paper proposes a gray-box approach to modeling and simulation of photo-voltaic modules. The process of building a gray-box model is split into two components (known, and unknown or partially unknown). The former is based on physical principles while the latter relies on functional approximator and data-based modeling. In this paper, artificial neural networks were used to construct the functional approximator. Compared to the standard mathematical model of photovoltaic module which involves the three input variables - solar irradiance, ambient temperature, and wind speed- a gray-box model allows the use of additional input environmental variables, such as wind direction, atmospheric pressure, and humidity. In order to improve the accuracy of the gray-box model, we have proposed two criteria for the classification of the daily input/output data whereby the former determines the season while the latter classifies days into sunny and cloudy. The accuracy of this model is verified on the real-life photo-voltaic generator, by comparing with single-diode mathematical model and artificial neural networks model towards measured output power data.