Practical Battery Health Monitoring using Uncertainty-Aware Bayesian Neural Network

TL;DR

The paper tackles practical battery health monitoring by integrating uncertainty quantification into end-of-life prediction using Bayesian neural networks, enabling distributions rather than single-point forecasts. It proposes an uncertainty-aware pipeline that processes sensor-derived features to predict the EoL cycle $c^{\text{EoL}}$ with a 95% confidence interval, and it demonstrates how predictions improve in accuracy and certainty as more cycle data become available. The study leverages the Stanford-MIT battery aging dataset, introduces a DenseFlipout-based BNN with an IndependentNormal output, and shows competitive MAE/MAPE alongside meaningful uncertainty measures; results indicate a mean absolute error of around 13.9% at later stages and substantial certainty gains (e.g., 66% CI improvement from cycle 100 to 400). This work has practical implications for battery management systems, supporting safer, more reliable, and cost-effective deployment of electric-vehicle batteries by informing maintenance, replacement, and repurposing decisions with quantified risk.

Abstract

Battery health monitoring and prediction are critically important in the era of electric mobility with a huge impact on safety, sustainability, and economic aspects. Existing research often focuses on prediction accuracy but tends to neglect practical factors that may hinder the technology's deployment in real-world applications. In this paper, we address these practical considerations and develop models based on the Bayesian neural network for predicting battery end-of-life. Our models use sensor data related to battery health and apply distributions, rather than single-point, for each parameter of the models. This allows the models to capture the inherent randomness and uncertainty of battery health, which leads to not only accurate predictions but also quantifiable uncertainty. We conducted an experimental study and demonstrated the effectiveness of our proposed models, with a prediction error rate averaging 13.9%, and as low as 2.9% for certain tested batteries. Additionally, all predictions include quantifiable certainty, which improved by 66% from the initial to the mid-life stage of the battery. This research has practical values for battery technologies and contributes to accelerating the technology adoption in the industry.

Figures (3)

• Figure 1: An illustration of the system architecture of ML-based battery EoL prediction. Using the battery's health data about charging and discharging, an ML model predicts the battery's expected EoL together with a quantifiable certainty as a 95% confidence interval (CI). The prediction results are used by the BMS to offer early EoL alerts and other health-related services.
• Figure 2: A case study of Bnn's performance on a specific battery cell showing prediction distribution at cycles 100, 200, 300, and 400. In each sub-figure, the battery's actual EoL is shown as a dashed vertical line at cycle 788. Both the lower limit and upper limit of the 95% CI in each sub-figure are shown in dotted vertical lines. Bnn's EoL prediction achieves improved accuracy and certainty with more cycles of information.
• Figure 3: A comparison study of Bnn and its counterparts with the same data but different ML models for MAE in Fig. \ref{['fig:CompMAE']} and MAPE in Fig. \ref{['fig:CompMAPE']}. Bnn achieves competitive prediction accuracy besides its quantifiable uncertainty.