Estimating Battery State of Health using Machine Learning

Abstract

The share of energy consumption in the transportation sector is projected to increase at an annual average rate of 1.4% up to 2040. This is primarily due to a transition towards electric vehicles (EVs) from internal combustion engine- based modes of transportation. Batteries are the most crucial component in EVs, constituting a significant share of the price of the vehicle. With usage, batteries degrade, thereby, limiting their ability to store energy which adversely impacts the driving range offered by EVs. Therefore, the need is to study the deterioration of batteries in electric means of transportation. We have created data-driven models to monitor battery health, predict the deterioration in batteries and give insights to the EV owners to make better decisions. The dataset used in this study is published by Sandia National Labs (SNL). It is a result of experiments performed on NMC cells. We present a comparison of three models - multiple linear regression, support vector regression, and artificial neural network for battery health monitoring with mean average percentage error (MAPE) of 1.99, 0.74, and 0.72 respectively.