On-line neural network training for maximum power point tracking of PV power plant

Abstract

The paper presents a new method to achieve maximum power point tracking (MPPT) for practical grid-connected PV panels. The method employs the radial basis function neural network (RBFNN) to predict the PV plant's maximum power points corresponding to different weather conditions. The RBFNN model can be trained on-line, autonomously, using a simplified genetic algorithm (SGA). The method has been verified by modelling the MPP points for practical PV panels located in Southampton and Leeds, respectively. The Leeds model has been used to control a prototype PV-grid-connected power generation system. The on-line RBFNN training scheme is discussed in detail and experimental results are presented which compared favourably with the conventional perturbation and observation (P&O) technique.