Accurate State of Health Estimation of Lithium-ion Batteries: An Efficient ResNet-LSTM Approach

Accurate estimation of the state of health (SoH) of lithium-ion battery (LIB) application systems is a critical concern in the domain of electric vehicles (EVs). Precise SoH holds significance due to its direct impact on the performance, longevity, and safety of LIBs. This paper proposes an efficient deep learning (DL) method for SoH estimation. The proposed DL model incorporates a residual neural network (ResNet) with long short-term memory (LSTM), extracting spatial-temporal features from LIB data. The ResNet-LSTM approach ensures high accuracy in SoH prediction. Extensive experimental results derived from real-world LIBs are presented, demonstrating the model's superiority and competitiveness in score measures when compared to existing SoH estimation models. The key contributions of this paper include (1) the simulation of a highly effective model explicitly tailored for SoH estimation in LIBs for EVs; (2) empirical verification of the effectiveness and superiority of the proposed model on the widely used NASA battery dataset. Simulation results are carried out to demonstrate the generalization and feasibility of the ResNet-LSTM model in SOH estimation with an overall root-mean-squared error of 0.69%.