Rastreamento de máxima potência fotovoltaica através de redes neurais artificiais perceptron multicamadas

Abstract

This work presents an alternative to conventional maximum power dotting (MPPT) tracking methods through the use of artificial neural networks (RNA) as a function approximation. The present work develops a technique that allows to improve the performance of the photovoltaic system both in relation to the steady state error and in the dynamic response of the extracted power in the photovoltaic array. The developed method uses a multi-layered Perceptron Artificial Neural Network (MLP) as an approximation of the maximum power voltage, by comparing this reference with the current operating voltage of the photovoltaic system (𝑉𝑝𝑣), an error signal is generated allowing the definition of an optimal increment of the duty cycle step. The results obtained demonstrate a performance far superior to the conventional classical Perturb and Observe (𝑃&𝑂) method. First, the current context and implications of MPPT tracing will be presented. A state-of-the-art literature review will then be carried out in conventional MPPT methods and using RNA. Also presented are the methodology applied in the study and, finally, the results obtained with the proposed system.