Data-driven Modelling for EV Battery State of Health Estimation using SFS-PCA Learning

Transportation electrification is a crucial pathway for engineering net-zero, and Lithium-ion batteries have been widely used for electrified vehicles (EV). Estimating battery state-of-health (SoH) is a critical task in EV development, and advanced modeling methods are required. This paper studied data-driven modeling for SoH estimation using deep learning. By incorporating Sequential Feature Selection (SFS) with Principal Component Analysis (PCA), a new deep learning method, SFS-PCA learning, is proposed for battery SoH estimation with three stages. In the first stage, the battery degradation features (e.g., voltage, current, and temperature) were normalized and selected by the SFS module based on min-max feature scaling and linear regression to minimize the number of input variables for deep learning. In the second stage, the PCA module transformed the input variables and minimized the estimation model’s computational burden by removing redundant feature information. In the third state, a deep neural network model is developed and trained with the selected and transformed input variables and NASA battery testing data. Using deep learning models without feature selection and PCA transformation and other machine learning models such as SVR and Random Forest as baseline methods, SoH prediction performance (e.g., RMSE) was compared and evaluated. The study suggested that the proposed SFS-PCA-Deep Learning method can reduce the RMSE by 54% at a high R
²
level of 0.96.