Efficient Low-Cost Method For The Estimation Of Clouds Shading Rate on PV Farms - Real-Time Reconfiguration Application

Abstract

In large photovoltaic plants (PV farms), the partial shading phenomenon due to clouds leads to a decrease in the total power of the installation, and consequently a decrease in profitability. For improving the plant yield although partially shaded, a reconfiguration of the panel's connection is performed. Several reconfiguration algorithms are published, and a switch matrix is controlled providing panels interconnectivity. However, these algorithms are based on measurements made on each panel of the farm providing the solar radiation on it, its output voltage, and current. This conventional method needs a complex and expensive installation of a great number of instruments, it requires also regular maintenance. This research aims to find a low-cost and spacesaving reconfiguration method using the least possible measuring devices. A feed-forward neural network is developed to estimate the shading rate of each PV zone using meteorological data and sun position. It is found that training a regression model responds better to the problem. Also, training a feed-forward neural network with the Levenberg-Marquardt algorithm proves to be efficient in terms of stability, speed, and convergence with a small error.