An Ensemble Convolutional Neural Network Based Transfer Learning Framework for Achievable Lithium-Ion Battery Health Perception in Flexible Charging Procedure

Lithium-ion batteries may be charged in various state of charge regions due to the electric vehicle users’ charging behavior, which increases the difficulty of an accurate state of health (SOH) estimation. The existing methods tend to use a fixed charging voltage for extracting the features for SOH estimation, which limits their accuracy and generality. To address this issue, this article proposes an ensemble convolutional neural network-based transfer learning framework for battery SOH estimation in arbitrary charging voltage. First, the random charging interval is divided into a group of consistent voltage segments. The incremental capacity and the voltage of corresponding segments are chosen as the health indicators. Then, a domain adaptive neural network (DaNN) transfers the features to arbitrary charging voltage. To improve the estimation accuracy, an ensemble learning model with an adaptive correlation vector machine and blending optimization is designed. The model's core lies in generating multiple combinations of optimal training subsets of DaNN models based on different voltage segments. Finally, a relevance vector machine is used for SOH estimation. To verify the model, aging tests are conducted on four lithium-ion batteries. The experimental results show that the proposed method can achieve a high estimation accuracy for arbitrary charging voltage.