Enhancing Lithium-Ion Battery Management with Advanced Kalman Filter Tuning

Enhancing Lithium-Ion Battery Management with Advanced Kalman Filter Tuning summarises the work conducted by Jasper Knox, Mark Blyth and Alastair Hales (all University of Bristol).

Lithium-ion batteries are integral to modern electric vehicle development, requiring advanced battery management systems (BMS) for effective battery pack operation. A critical task for these systems is accurately estimating state of charge (SOC) in real-time, which directly impacts the vehicle’s driving range and battery longevity.

Overcoming Traditional SOC Estimation Challenges

Basic SOC estimation methods such as Coulomb counting are difficult to implement practically in BMS applications. Instead, predictions of SOC are performed using algorithms such as the extended Kalman filter. Kalman filters are used to integrate battery models with real-time measurements of voltage, current, and temperature to provide a more accurate estimation of SOC than could be obtained either through measurement or modelling individually. However, appropriate tuning of Kalman filter parameters remains a difficult task and inadequately tuned filters may produce unstable results. A recent study proposed an adaptive Kalman filtering scheme for estimating SOC in battery models with hysteresis, where uncertainty in filter parameters is used to regulate SOC estimation at each timestep. We provide experimental validation of this method and demonstrate that a well-parameterised hysteresis model provides very high SOC estimation performance when paired with an adaptive extended Kalman filter.

Model Development

We tested a range of equivalent circuit models (ECM) with the extended Kalman filter.  These simple battery models are easy to parameterise, making them ideally suited for real-time operation in BMS applications. A hysteretic ECM was compared to the industry-standard Thévenin model. The hysteretic ECM introduces an additional hysteresis voltage component, allowing for more accurate representation of voltage dynamics during kinetic loading and rest periods. Model parameters were identified from experimental data using the galvanostatic intermittent titration technique (GITT). Results demonstrate that a well-parameterised

Advanced Kalman Filter Tuning comparison of errors
Figure 1: The mean RMS voltage prediction error for standard (ECM) and hysteresis (ECMh) models, evaluated against the Worldwide Harmonised Light Vehicle Test Procedure (WLTP). Models with the suffix “opt” have optimised parameters. Error bars report the maximum and minimum RMS voltage prediction error for each model across all cells.

hysteresis model reduces RMS voltage prediction error by approximately 50 % on relevant test data compared to the model without hysteresis, as shown in Figure 1.

Filter Performance

We compared SOC estimation performance across all ECMs and Kalman filter schemes. The adaptive extended Kalman filter improved SOC estimation performance (up to 85%) compared with constant tuning for all cell models. Enhancements to the Kalman filter yielded greater performance benefits than improving the underlying model with hysteresis components or parameter optimisation. These findings should be validated on cell chemistries beyond NMC, particularly those with substantial hysteresis in their open-circuit voltage, such as LFP cells.

Conclusions and Future Work

In this study, equivalent circuit models were parameterised, with hysteresis models identified as the most effective in reducing RMS voltage prediction error. All models were coupled with an extended Kalman filter to estimate SOC, comparing constant and adaptive filter tuning performance. Critically for BMS designers, improving the Kalman filter was shown to be more effective than refining the underlying battery models. Future research should validate these findings across various cell chemistries and accurately characterise uncertainty in filter parameters, which is key for the success of an adaptive filtering scheme. Incorporating explicit temperature dependency in cell models and associated Kalman filters is required for practical implementation. Additionally, since typical battery packs consist of many cells, scaling adaptive filtering methods to pack-level SOC estimation is essential.

Advanced Kalman Filter Tuning - paper

This article summarises the work conducted by Jasper Knox, Mark Blyth and Alastair Hales (all University of Bristol). The researchers are part of the Electrical Energy Management Group, a research group within the Faculty of Engineering at the University of Bristol. The research group contains 15 academics and 50+ postgraduate researchers, delivering world-leading research across power electronic, electrical machines and energy systems.

To read the full open access paper:

Knox, J., Blyth, M. & Hales, A. Advancing state estimation for lithium-ion batteries with hysteresis through systematic extended Kalman filter tuning. Sci Rep 14, 12472 (2024). https://doi.org/10.1038/s41598-024-61596-0

Leave a Comment