An EM approach for dynamic battery management systems

In this paper we propose an expectation maximization (EM) type algorithm for online system identification and tracking of prognostic states as applied to battery management systems (BMS). The objective of BMS is to adaptively estimate state of charge (SOC) - a crucial battery state information. We find that the existing approaches in the literature need enhancement because (i) they lack battery equivalent models that accurately model the actual physical and chemical properties of the batteries, and (ii) they fail to utilize powerful state and parameter estimation techniques for system identification and tracking. It is often noticed throughout the literature that (ii) is a precursor to (i). In this paper, first we model the dynamic equivalent model of batteries as a series of m parallel RC circuits and derive the relationship between the time-varying battery states and the current, voltage output observations as a non-linear state space model. Then, we derive an expectation maximization (EM) type algorithm for identification of the so-derived statespace model and for the adaptive tracking of SOC. Finally we discuss the performance evaluation of the proposed algorithm through simulation and by testing them on experimental data obtained from Li-ion based cellphone batteries.