A likelihood-partitioned Bayesian framework for lithium sulfur battery state discharging of charge estimation

Abstract

Lithium sulfur (Li-S) batteries are promising energy storage devices and alternative to lithium-Ion (Li-Ion) batteries in electric grid and vehicle applications. However, compared to Li-Ion, the discharge voltage of Li-S is much complex and nonlinear. This results a challenging state of charge (SoC) estimation problem while Li-S is discharging. For such a problem, the traditional extended Kalman filter fails to provide accurate SoC. Therefore, this paper proposes a novel likelihood partitioned Bayesian filtering (LPBF) framework and its linearized version for SoC estimation of discharging Li-S battery cell. Though both traditional EKF and linearized LPBF use a prediction error minimization based equivalent circuit network (ECN) parameterization, the LPBF uses a partitioned ECN parameterization. The portioned models result two likelihoods, whereas the EKF uses a single state-space model throughout discharge from 100 percent SoC to zero SoC. With experiment data obtained at two different temperature conditions, numerical simulation results compare both EKF and linearized LPBF based SoC estimators. Simulation results show that the LPBF's accuracy is impressive, about 97 percent, for considered dynamic load current, operating temperature and uncertain initial SoC conditions.