Estimativa do estado de carga de baterias utilizando redes neurais artificiais

Abstract

With the growing demand for electric vehicles, the development of intelligent systems capable of accurately monitoring the functionality and longevity of batteries has become essential, as they are among the most critical and costly components of these automobiles. Proper management of the State of Charge (SOC) is crucial to ensure battery performance, safety, and lifespan. In this context, the present work aims to estimate the SOC of batteries from the Renault Kangoo, an electric vehicle with approximately 10 years of use, using machine learning techniques. Three distinct Artificial Neural Network (ANN) architectures were implemented and evaluated: the Multilayer Perceptron (MLP), the Long Short-Term Memory network (LSTM), and the Extreme Learning Machine (ELM). The methodology involved collecting, preprocessing, and analyzing real-world operational data from the vehicle to train and validate the proposed models. The best performance was achieved by the ELM and LSTM networks, with mean squared errors of 0.000023 and 0.000048, respectively, demonstrating high accuracy in SOC prediction. The results highlight the potential of neural networks for intelligent battery monitoring in vehicular applications.