State-of-Charge Estimation of Lithium-Ion Batteries Using a Computationally Efficient Particle Filter

Accurate state of charge (SOC) estimation is essential for battery management systems (BMS) in electric vehicles (EVs), especially under nonlinear dynamics, non-Gaussian noise, and highly dynamic operating conditions. Particle filters (PFs) are well suited to these challenges; however, conventional implementations such as the sampling importance resampling particle filter (SIRPF) often suffer from weight degeneracy, particle impoverishment, and high computational cost, which limit their suitability for real-time embedded applications. To overcome these issues, this paper proposes a percentile-based resampling (PBR) strategy integrated into the PF framework, resulting in a percentile-based particle filter (PBPF) that reformulates the resampling mechanism by retaining percentile-ranked informative particles and regulating particle replication according to their importance weights. The proposed PBPF is evaluated using the widely adopted LA92 driving cycle at $40^{\circ } \mathrm {C}$ and further validated under additional operating temperatures of $25^{\circ } \mathrm {C}$ and $0^{\circ } \mathrm {C}$ , as well as the NASA B0005 battery at the 10th discharge cycle. Comparative results against the EKF, AEKF, conventional SIRPF, and extended particle filter (EPF) demonstrate that the proposed PBPF consistently achieves superior SOC and terminal voltage estimation performance. Specifically, the proposed PBPF achieves the best SOC RMSE values of 1.308%, 0.405%, and 0.236% for the LA92 driving cycle at $40^{\circ } \mathrm {C}$ , $25^{\circ } \mathrm {C}$ , and $0^{\circ } \mathrm {C}$ , respectively, and 0.07% for the NASA B0005 battery at the 10th discharge cycle. These correspond to the highest overall SOC improvements of 26.19%, 74.79%, 83.57%, and 98.66% over the EKF/dual EKF baseline, respectively. These results demonstrate that the proposed PBPF provides a favorable accuracy–complexity trade-off, making it well suited for real-time embedded BMS applications.