Machine Learning on Electric Vehicle Lithium-Ion Battery Cycling Data for Health Estimation

Lithium-ion batteries, widely used in electric vehicles and consumer electronics, are essential to modern energy storage systems. While their primary electrochemical reactions facilitate efficient energy storage and release, complex and often immeasurable side reactions contribute to gradual degradation, reducing battery capacity over time. This degradation shortens battery lifespan and poses risks of system failures, property damage, and personal injury. Accurately predicting a battery’s State of Health (SOH) is critical for ensuring safety, optimizing performance, and extending battery longevity. Given the inherent complexity and ambiguous nature of battery degradation mechanisms, machine learning (ML) and deep learning (DL) techniques, particularly neural networks, offer powerful solutions. In this study, we trained multiple models on extensive battery cycling data and applied feature engineering techniques to enhance prediction accuracy. Specifically, we computed the State of Charge (SOC) and the derivatives of voltage and current (dV/dt and dI/dt) to capture dynamic cycling behavior, incorporating them as additional input features. Our approach includes traditional linear and polynomial regression models as well as neural networks to improve SOH estimation. By highlighting the limitations of traditional ML techniques, we demonstrate the advantages of modern DL approaches in accurately predicting SOH. Accurate SOH prediction not only mitigates safety risks but also promotes the sustainable use of lithium-ion batteries, offering significant environmental benefits by extending battery life and reducing electronic waste.

Recommended citation: Your Name, You. (2010). "Paper Title Number 2." Journal 1. 1(2).
Download Paper | Download Slides