Multicell state estimation using variation based sequential Monte Carlo filter for automotive battery packs

• A variation based framework for multicell state estimation is studied.
• Battery mean SOC is estimated in form of a probability density function.
• Cell-to-cell variation is reconstructed via recursive least squares method.
• Comparison of computational cost of different estimation approaches is given.
• Estimation performance are evaluated by simulation and experiment.

Accurate state monitoring is required for the high performance of battery management systems (BMS) in electric vehicles. By using model-based observation methods, state estimation of a single cell can be achieved with non-linear filtering algorithms e.g. Kalman filtering and Particle filtering. Considering the limited computational capability of a BMS and its real-time constraint, duplication of this approach to a multicell system is very time consuming and can hardly be implemented for a large number of cells in a battery pack. Several possible solutions have been reported in recent years. In this work, an extended two-step estimation approach is studied. At first, the mean value of the battery state of charge is determined in the form of a probability density function (PDF). Secondly, the intrinsic variations in cell SOC and resistance are identified simultaneously in an extended framework using a recursive least squares (RLS) algorithm. The on-board reliability and estimation accuracy of the proposed method is validated by experiment and simulation using an NMC/graphite battery module.