Battery State of Health Estimation Based on Energy Features and ResNet-SVR Model

Abstract

The accurate estimation of battery state of health (SOH) is crucial for monitoring battery health status and alerting users to replace the degraded batteries. Energy, as a comprehensive feature integrating current and voltage, can effectively reflect the amount of stored battery energy, and graphical energy features contain more information. In this study, a hybrid ResNet-SVR model based on energy features is proposed to estimate the battery's SOH. The sequential energy features are first constructed during the constant current charging process. Then they are transformed into a two-dimensional graphical structure through the sliding window as model inputs, which shows significant differences between cycles. The hybrid ResNet-SVR model uses ResNet to extract relevant information from the graphical energy data and SVR for regression prediction. It combines the advantages of a convolutional neural network for feature extraction and SVR for regression prediction to achieve better results. The proposed method is validated on two battery datasets and achieves satisfactory results in battery estimation, with an average mean absolute error of 0.0072 and a mean absolute percentage error of 0.83%. Overall, our findings highlight the potential of the hybrid ResNet-SVR model with energy features for accurate battery SOH estimation, which significantly enhances the performance of the battery management system in various applications.