A multiscale data-driven framework for lithium-ion battery on-line state estimation

Abstract

Lithium-ion battery is widely applied in lots of different industrial fields including spacecraft, electrical vehicle (EV), renewable energy systems, etc. Accurate lithium-ion battery state estimation including state of charge (SOC) and state of health (SOH) is meaningful for both battery operation reliability increase and battery maintenance cost decrease. Traditional SOC estimation methods focus on establishing a mathematical or equivalent circuit model to make the SOC strongly linked to the cell voltage and current. However, in real applications, the complex operating condition brings great challenges in on-line parameter identification and model updating. On the other hands, on-line SOH estimation also suffers from the lack of accuracy since the battery actual capacity and impedance are neither measurable under real operating conditions. To address these challenging issues in both SOC and SOH estimation, this paper proposed a multiscale data-driven framework for lithium-ion battery on-line estimation. A parameter-free model is established for SOC estimation. Least square support vector machine (LS-SVM) is introduced into the framework for model training. A direct mapping model is also trained by LS-SVM for SOH estimation based on two health indicators extracted from online measurable parameters including voltage, current and time interval. SOH estimation results are also applied as the feedback for SOC estimation to further improve the estimation accuracy. Experimental results verify the effectiveness and robustness of the proposed framework.