A comparative study evaluates four neural architectures—MLP, ResNet, U-Net, and FNO—as autoregressive predictors of internal battery states using the Doyle-Fuller-Newman model. The U-Net achieves a mean final-step nRMSE of 3% across all state variables and provides a 5.38x speed-up over numerical solvers, demonstrating the importance of spatial inductive bias in surrogate performance.
Comparative Study of Neural Surrogates for Battery State Prediction
from English