A data-selective strategy for online lithium-ion battery parameter estimation based on vehicle trip data

Accurate online battery parameter estimation methods are highly important in Electric Vehicle (EV) applications for ensuring battery pack performance and safety under dynamic conditions. To this end, it is necessary for the estimation method to provide low-bias and low-variance estimates as efficiently as possible. In previous literature, data selection has been shown to reduce the overall data use of the estimation method, while enhancing the resulting performance, by only updating parameter estimates on measurement data with high informational content. In this paper, data selection is extended to the errors-in-variables problem by classifying informative data segments based on the derived Fisher information under the assumption of noise in both the input and output data. High-information battery data segments are then identified from the speed and acceleration of the EV, based on a trained time-series classifier, and used to update battery parameters using Total Least Squares (TLS). The resulting parameters are subsequently used in estimating the State-of-Charge (SOC) and State-of-Power (SOP) of the EV. Both simulation and experimental results show that the proposed data-selective strategy significantly reduces the bias of two out of three parameters in an auto-regressive with exogenous input (ARX) model compared to Recursive Least Squares (RLS) in a high-noise environment, while keeping the variance low. Additionally, the data-selective strategy is able to significantly outperform RLS in SOC and SOP estimation using less than 10% of the available data. A computational analysis further demonstrated that the data-selective strategy reduces the total amount of computations in the distributed boards by 97.25%–98.48% relative to RLS.