No abstract

In this study, a vehicle localization technique was employed to determine the required quantities in the identification of battery models by considering the behavior of multiple batteries instead of data from a single battery. In previous studies, a plant (e.g., a battery, motor, super-capacitor, or fuel cell) was identified based on a single piece of data. However, such an approach is disadvantageous in that it neglects the effect of process and measurement noise and assumes that the parameters obtained using data from a single plant are identical for all plants of the same type. First, deterministic parameter estimation (DPE), particle swarm optimization (PSO), and teaching-learning-based optimization (TLBO) were initially applied to estimate the battery model parameters using data from a single battery. Second, a fusion-based approach was used to address the process and measurement noise problems through an adaptive unscented Kalman filter algorithm. With this approach, maximum likelihood estimation was employed to fuse multiple-battery data streams to enable the DPE, PSO, and TLBO to recalculate the model parameters based on filtered and fused quantities. A comparison between the experimental results and model outputs obtained using the aforementioned methods for parameter estimation indicated that the proposed multiple-battery approach enhances the accuracy of several identification methods. In contrast, it requires a high computational effort. INDEX TERMS universal adaptive stabilizer (UAS), particle swarm optimization (PSO), teachinglearning-based optimization (TLBO), unscented Kalman filter (UKF), and maximum likelihood estimation (MLE).