Predição da descarga e capacidade de baterias li-on utilizadas em vants

Abstract

This research aims to perform a prediction study of the capacity and discharge of lithium ion batteries (Li-Ion) from a set of a NASA database. These batteries are commonly used in UAVs (Unmanned Aerial Vehicles) because it presents characteristics such as lightness, high energy density, large number of discharge cycles, among others. The prediction methods are used in the form of supervised learning, in which algorithms are fed part of the NASA database’s inputs and outputs. The methods used in this study were: Non-recursive least squares, Discrete Kalman Filter, Artificial Neural Network with ELM (Extreme Learning Machine) training, Particle Filter and Unscented Kalman Filter. In order to make a comparative analysis of these, RMSE (Root-Mean-Square Error), R2 (correlation coefficient) and computational cost were used. In addition, PHM (Prognostics and Health Management) techniques were used alongside the aforementioned ones, utilizing a threshold-based failure methodology as well as prediction barriers for the capacity with 95% confidence intervals.