Design, implementation and comparison of several neural perturb and observe MPPT methods for photovoltaic systems

Abstract

In the present article, several Genetic Algorithms (GAs) neural controllers MPPT strategies are used to apply to a voltage step-up converter driven by a PV stand-alone system. A new kind of control structures based on different input variables, combine three categories of techniques to develop an adaptive MPPT approach: the Perturb and Observe P&O technique is associated to a Multilayer Perceptron (MLP) using GAs as learning suggestion abilities. Used in order to optimize the controller efficiency by driving the network through the use of new information in synaptic connections to get an optimal learning rule whatever the changing weather and load conditions. Based on environmental sensors an experimental data have been collected to settle the learning sets for the learning algorithm of the MLP. In consequence, the performances of the proposed GA neural controllers inserted in a complete MPPT strategy have been validated with simulation tests using the Matlab/Simulink environment. Good result is confirmed and compared inside a test bench based on real PV system with boost converter and resistive load piloted by the dSPACE 1104 card. The superior characteristics of the electrical structure using GAs neural controller is affirmed in terms of performance assessment covers overshoot, robustness, transient/steady-state performance, time response and oscillations.