Health Prognosis of Lithium-Ion Batteries using CNN-SVR Fusion Model for Dynamic Discharge

The state of health (SoH) estimation of lithium-ion batteries (LIBs) plays a pivotal role in safe and reliable operation of an electric vehicle (EV). Numerous methods have been proposed in the recent times for LIB's SoH diagnostics and prognostics. The accuracy in terminal voltage estimation during charging and discharging is equally crucial to extract the health indicators (HIs) required for the SoH estimation. However, in practical driving scenario, LIB charges and discharges dynamically resulting in incomplete capture of voltage curve details. Therefore, this paper proposes an SoH estimation framework depending on terminal voltage extraction based fusion model. Firstly, the discharge terminal voltage is reconstructed based on importance sampling to obtain voltage-time (V-t) curve for batteries operating under variable temperatures. Secondly, the HIs related to SoH are extracted and their co-relation with SoH is analyzed. Furthermore, a fusion model based on convolutional neural network (CNN) and support vector regression (SVR) is established. The results show that the performance evaluation indicators of the proposed framework are superior as compared to SVR and CNN alone, with root mean square error (RMSE) of less than 0.00416 and mean absolute error (MAE) less than 0.00315 for all batteries.