Volume 37: New Energy, New Ecology and New Environment

CNN-based reconstruction of capacity degradation trajectory of lithium-ion batteries Ertao Lei; Li Jin; Shuowei Li; Jingcai Du; Caiping Zhang

Abstract

Accurate lithium-ion battery health estimation is crucial to ensure the safe and stable operation of energy storage battery systems. To address the problem of inaccurate battery state of health (SOH) estimation due to low sampling frequency and few stored data in the energy storage battery system, this paper proposes a battery capacity degradation trajectory reconstruction method based on convolutional neural network (CNN). Firstly, the battery capacity increment curves are analyzed to select the voltage segments with obvious differentiation for various degradation states of batteries. Secondly, the selected voltage-capacity segments in the first 30 cycles of batteries are input to a 3-layer CNN. Finally, the life-span capacity degradation curves are directly reconstructed without artificially feature selection and any voltage-capacity data after 30 cycles. The results show that the method has a high accuracy of capacity reconstruction with a mean absolute percentage error (MAPE) within 0.7%.

Keywords Lithium-ion battery, convolutional neural network, capacity trajectory reconstruction