Application of radial basis function networks for solar-array modelling and maximum power-point prediction

A neural-network-based approach for solar array modelling is presented. The logic hidden unit of the proposed network consists of a set of nonlinear radial basis functions (RBFs) which are connected directly to the input vector. The links between hidden and output units are linear. The model can be trained using a random set of data collected from a real photovoltaic (PV) plant. The training procedures are fast and the accuracy of the trained models is comparable with that of the conventional model. The principle and training procedures of the RBF-network modelling when applied to emulate the I/V characteristics of PV arrays are discussed. Simulation results of the trained RBF networks for modelling a PV array and predicting the maximum power points of a real PV panel are presented.