Power Extraction of a Solar PV Module using Photocurrent Based RBF Neural Network

Abstract

Modelling of I-V characteristics of solar PV is vital to predict and track the maximum power. The total current output from a panel consists of a photocurrent dependent on the solar irradiation and the temperature, the saturation current and the diode current. This relationship is inherently nonlinear and involves several parameters required to derive using complex data analysis as they are unavailable in most manufacturers’ datasheets. Furthermore, these parameters depend highly on the material properties and construction of the PV modules and the environmental conditions. Radial basis function neural networks (RBF-NNs) are special feedforward neural networks with universal approximation properties. An RBF-NN structure is simple and compact, and its learning and training are fast. The main objective of this study is to model the nonlinear behaviour of the single-diode solar PV model using the measured data. Therefore, RBF-NN is a suitable candidate for handling the model nonlinearity. The PV module’s output current is estimated by combining the calculated PV module output using photocurrent and the RBF-NN output. The combined photocurrent based NN model is trained until the mean squared error is less than 0.0001W. The combined RBF-NN was trained and validated using data sets with 1 min intervals obtained from the open-rack PV modules installed on a flat roof test station. This data set contains temperature, solar irradiation, current, voltage and power output data. The RBF-NN model gives significantly less error compared to the error of the measured output from the module. The maximum error of the RBF-NN is 0.0582W, and that for measured output is 19.5142W. Therefore, the proposed model can be used to track the maximum power output of the solar PV module and compare existing maximum power point tracking algorithms using measured temperature and irradiance data accurately.