Kalman Filter Tuning Using Multi-Objective Genetic Algorithm for State and Parameter Estimation of Lithium-Ion Cells

Abstract

To ensure a reliable and safe operation of battery systems in various applications, the system’s internal states must be observed with high accuracy. Hereby, the Kalman filter is a frequently used and well-known tool to estimate the states and model parameters of a lithium-ion cell. A strong requirement is the selection of a suitable model and a reasonable initialization, otherwise the algorithm’s estimation might be insufficient. Especially the process noise parametrization poses a difficult task, since it is an abstract parameter and often optimized by an arbitrary trial-and-error principle. In this work, a traceable procedure based on the genetic algorithm is introduced to determine the process noise offline considering the estimation error and filter consistency. Hereby, the parameters found are independent of the researcher’s experience. Results are validated with a simulative and experimental study, using an NCA/graphite lithium-ion cell. After the transient phase, the estimation error of the state-of-charge is lower than 0.6% and for internal resistance smaller than 4mΩ while the corresponding estimated covariances fit the error well.