Predicting the state of charge of lithium ion battery in E Vehicles using Box-Jenkins combined Artificial Neural Network Model

Abstract This manuscript used artificial neural networks to predict the state of charge of lithium-ion batteries in electric vehicles. For this, a hybrid model that combined Box–Jenkins and artificial neural network techniques was used. The original Auto-Regressive Moving Average (ARMA) model was developed in three stages: finding the best fit using Auto-Correlation Function (ACRF) and Partial Auto-Correlation Function (PACRF) in the first stage, parameter estimation in stage two & verification in stage three using the Ljung-box technique. For the purpose of estimating the dynamic system response, the second model was developed using a Multi-Layer Perceptron (MLP) network with feedforward backpropagation. Prediction accuracy was significantly increased by the combination model, which integrated non-linear Artificial Neural Network model with linear Auto-Regressive Moving Average model. Notably, the state of charge of lithium-ion batteries in electric vehicles could be accurately predicted using a four-parameter model that included Charge Rate, Voltage, Depth of Discharge, and Energy Density.