Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Coupling Neural Networks and Physics Equations For Li-Ion Battery State-of-Charge Prediction

Giovanni Pollo, Alessio Burrello, Enrico Macii, Massimo Poncino, Sara Vinco, Daniele Jahier Pagliari

TL;DR

The paper tackles the challenge of predicting the battery State of Charge (SoC) both now and across future horizons for smarter power management. It introduces a two-branch neural network where one branch estimates the current SoC from measurable signals and a second branch forecasts future SoC under specified workloads; training is augmented with a physics-informed loss based on Coulomb counting to regularize future predictions across multiple horizons. Evaluations on Sandia and LG datasets show the approach achieves competitive accuracy with significantly fewer parameters and computations than state-of-the-art methods, while improving generalization to unseen horizons. The results demonstrate strong potential for multi-horizon SoC forecasting in battery management systems and onboard devices, with autoregressive full-discharge analysis illustrating practical benefits and limitations.

Abstract

Estimating the evolution of the battery's State of Charge (SoC) in response to its usage is critical for implementing effective power management policies and for ultimately improving the system's lifetime. Most existing estimation methods are either physics-based digital twins of the battery or data-driven models such as Neural Networks (NNs). In this work, we propose two new contributions in this domain. First, we introduce a novel NN architecture formed by two cascaded branches: one to predict the current SoC based on sensor readings, and one to estimate the SoC at a future time as a function of the load behavior. Second, we integrate battery dynamics equations into the training of our NN, merging the physics-based and data-driven approaches, to improve the models' generalization over variable prediction horizons. We validate our approach on two publicly accessible datasets, showing that our Physics-Informed Neural Networks (PINNs) outperform purely data-driven ones while also obtaining superior prediction accuracy with a smaller architecture with respect to the state-of-the-art.

Coupling Neural Networks and Physics Equations For Li-Ion Battery State-of-Charge Prediction

TL;DR

The paper tackles the challenge of predicting the battery State of Charge (SoC) both now and across future horizons for smarter power management. It introduces a two-branch neural network where one branch estimates the current SoC from measurable signals and a second branch forecasts future SoC under specified workloads; training is augmented with a physics-informed loss based on Coulomb counting to regularize future predictions across multiple horizons. Evaluations on Sandia and LG datasets show the approach achieves competitive accuracy with significantly fewer parameters and computations than state-of-the-art methods, while improving generalization to unseen horizons. The results demonstrate strong potential for multi-horizon SoC forecasting in battery management systems and onboard devices, with autoregressive full-discharge analysis illustrating practical benefits and limitations.

Abstract

Estimating the evolution of the battery's State of Charge (SoC) in response to its usage is critical for implementing effective power management policies and for ultimately improving the system's lifetime. Most existing estimation methods are either physics-based digital twins of the battery or data-driven models such as Neural Networks (NNs). In this work, we propose two new contributions in this domain. First, we introduce a novel NN architecture formed by two cascaded branches: one to predict the current SoC based on sensor readings, and one to estimate the SoC at a future time as a function of the load behavior. Second, we integrate battery dynamics equations into the training of our NN, merging the physics-based and data-driven approaches, to improve the models' generalization over variable prediction horizons. We validate our approach on two publicly accessible datasets, showing that our Physics-Informed Neural Networks (PINNs) outperform purely data-driven ones while also obtaining superior prediction accuracy with a smaller architecture with respect to the state-of-the-art.

Paper Structure

This paper contains 15 sections, 2 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Works
  3. Data-Driven SoC Models
  4. Methodology
  5. Network Architecture
  6. Training Scheme and Loss Function
  7. Datasets
  8. Sandia
  9. LG
  10. Experimental Results
  11. Results on the Sandia Dataset
  12. Results on the LG Dataset
  13. State-of-the-art Comparison
  14. Full discharge analysis
  15. Conclusions

Figures (5)

  • Figure 1: Proposed two-stage neural network architecture.
  • Figure 2: Multi-step autoregressive prediction with the proposed NN architecture.
  • Figure 3: Results on Sandia dataset with different physic loss
  • Figure 4: Results on LG dataset with different physic loss
  • Figure 5: Auto-regressive inference with our networks, on the four driving cycles of the LG dataset, at 25°C.