Universal Differential Equation Modeling of Lithium-Ion Battery State of Health
A hybrid degradation model in which the empirical calendar-aging and cycle-aging terms of a lithium-ion battery state-of-health ODE are replaced by neural networks, so the known Arrhenius physics is retained while the unknown dynamics are learned from data. Students learn how to embed a neural network inside a differential equation, train it against noisy synthetic and experimental capacity-fade data, and integrate the learned derivative forward to forecast state of health over multi-year horizons. The approach is contrasted with black-box time-series models and with purely empirical degradation fits that need chemistry-specific constants.
A Scientific Machine Learning Approach for Predicting and Forecasting Battery Degradation in
This paper presents a scientific machine learning methodology for predicting and forecasting battery degradation in electric vehicles over short and long time horizons. It directly addresses battery …