Modelo computacional para controle de sistemas de rastreamento solar de eixo duplo com utilização de técnicas de aprendizado profundo

Abstract

Solar energy is the cleanest and most abundant form of energy that can be obtained from the Sun. Solar panels convert this energy to generate solar energy, which can be used for various electrical purposes, in rural, industrial and residential areas. Maximum solar energy can be generated only when the Sun is perpendicular to the photovoltaic array, which can only be achieved for a few hours when using a fixed solar panel system. Over the years, different solar tracking systems have been proposed and developed, and some have been reviewed in the literature. However, these reviews did not consider two-axis solar tracking systems based on deep machine learning using recurrent neural networks. In this sense, the general objective of this work is to develop a computational model based on deep machine learning to control a two-axis solar tracking system to determine the best solar angle for positioning photovoltaic panels to improve the performance of photovoltaic panels in generation systems. energy through solar radiation. The model was developed using the architecture of recurrent neural networks of the Long Short-Term Memory type, the data used to create the training dataset and validation of the computational model were obtained from a system implemented in a simulation environment composed of four LDR´ s to capture solar radiation, a servo mechanism to control two motors in order to position the photovoltaic panels on the horizontal and vertical axes. The proposed model presents a prediction result for the positioning values of the two engines with an average error between the predictions and the actual values of 0.02, indicating that the model for controlling the two-axis solar tracking system based on deep machine learning is promising.