Rapid Assessment of Lithium-ion Batteries' SOH Based on the Segment of Charge/Discharge Voltage Curve Using Convolutional Neural Networks

Abstract

An essential part of lithium-ion battery management is the assessment of state of health (SOH), which is the key to accurate estimation of the state of charge and the remaining useful life. The majority of recent studies on SOH evaluation make use of the whole charging or discharging curves, first mining the variables with a strong correlation to SOH, and then building the projection to SOH using data-driven methods. However, this kind of method is difficult to achieve rapid assessment of SOH and the generalization of the mined features is poor. Therefore, we propose a rapid SOH assessment method based on the segment of charge/discharge voltage curve by using the powerful feature extraction ability of 1D-CNN. The results on the Oxford and NASA datasets demonstrate that the proposed method has a small prediction error and better generalization performance. In particular, the absolute error in SOH assessment for the Oxford dataset is below 5% with only 10s of voltage data.