Enhanced multi-state estimation methods for lithium-ion batteries considering temperature uncertainties

Abstract

Due to their high energy density and minimal emissions, lithium-ion batteries are frequently employed in electric vehicles (EVs). Accurate estimation of the micro-parameters, state of charge (SOC), and state of health (SOH) are a few primary monitoring functions of the battery management system (BMS) to increase the battery's efficiency and safety under various operating conditions. This paper proposes a SOC and SOH co-estimation method by adopting an ensemble empirical mode decomposition method with adaptive noise and an autoencoder (EEMDA) to extract, decompose, and reconstruct the full-scale charging voltage and current data for a dual extended Kalman filter (DEKF) with multi-parameter and time-scale updates for accurate estimation based on a variable forgetting factor limited memory recursive least squares (VFF-LMRLS) method. The VFF-LMRLS method is used to solve the data saturation phenomenon and identify the battery's characteristic micro-parameters based on a proposed dynamic migration second-order resistor-capacitor equivalent circuit model under different operating states. Battery tests are conducted at temperatures ranging from −10 to 50 °C under complex working conditions. Using the VFF-LMRLS method, the effects of different temperatures on the micro-parameters are discussed. The SOC and SOH results of the proposed EEMDA-DEKF method based on the dynamic migration battery model show that the mean absolute error and root mean square error metrics have the least values of 0.0233% and 0.0252%, which signify an optimal performance improvement of 93.26% and 93.66%, respectively, compared to the conventional DEKF method. Based on the experimental results and analyses, the proposed method has a high degree of accuracy and robustness, which makes it feasible for battery monitoring and prognostic BMS applications.