Battery Voltage Estimation Using NARX Recurrent Neural Network Model

Abstract

This work presents a prediction of battery terminal voltage in subsequent charging/discharging cycles. To estimate chosen signals the NARX (AutoRegressive with eXogenous input) model based on Recurrent Neural Network has been employed. A training and testing data were gathered at the laboratory test stand with the Lithium Iron Phosphate (LiFePO4) battery in different working conditions. Test stand research was conducted for 40 charging/discharging cycles. Furthermore, the paper presents the results of the identification of double RC model parameters for a specified state of charge level. As a result, the analysis of the proposed methodology has been discussed.