핵심 개념
Developing a PINN surrogate model for Li-ion batteries reduces computational resources and improves accuracy in parameter inference.
초록
This content discusses the implementation of a PINN surrogate model for Li-ion batteries, focusing on the single-particle model (SPM). The study aims to reduce computational resources needed for parameter inference and improve accuracy. The content covers the implementation, training strategies, regularization techniques, and hierarchical training for the PINN surrogate model.
-
Introduction
- Importance of electrochemical storage technology.
- Challenges in Li-ion battery technology.
-
PINN Surrogate Model
- Physics-informed neural network (PINN) surrogate for Li-ion battery models.
- Training strategies and regularization techniques.
-
Single-Particle Model
- Description of the SPM for Li-ion batteries.
- Linear vs. non-linear Butler-Volmer reaction.
-
Methods
- Implementation of PINN for battery models.
- Training regularization strategies.
-
Results
- Comparison of different neural net architectures.
- Effect of float precision on PINN accuracy.
- Hierarchical training approach for improved predictions.
통계
PINN surrogate can solve SPM 105 times faster and P2D model 106 times faster.
PINN training can be prone to instabilities.
PINN surrogate improves accuracy over traditional models.
인용구
"The surrogate model makes high-throughput techniques, such as Bayesian calibration, tractable to determine battery internal parameters."
"PINNs can rely on the governing equations of the system themselves to complement the lack of data."