Intelligent power conversion system management for photovoltaic generation

In this paper an intelligent management of a grid-connected PV system is proposed. The MPPT is based on the online estimation of the solar irradiance by the Growing Neural Gas (GNG) network. The PV system is composed of a DC/DC boost converter performing the MPPT and a single phase active rectifier controlled by a VOC algorithm for the connection to the grid. Each part of the PV system is controlled in a coordinated way with respect to the others, according to a general intelligent management strategy. The whole PV system, including the adopted neural-based MPPT, has been experimentally tested on a suitably devised test rig. The PV source is obtained by a power emulator to properly test the system under all possible operating conditions, including partial shading. A comparison between the proposed approach and a classical P&O technique has been done on a real irradiance profile on a daily scale, showing an increase of the generated power of 13%. The main drawback of the GNG-based MPPT is the need for a preliminary knowledge of the set of PV characteristics based on either a mathematical model or measured data, for the off line training of the GNG. Furthermore, the proposed MPPT exhibits a higher robustness with respect to the P&O under partial shading.