State of Health Estimation for Energy Storage Batteries Based on Feature Extraction and Neural Network Methods

Precise and reliable estimation of lithium-ion battery State of Health (SOH) is critical for battery management but is often hindered by complex degradation dynamics. This study introduces an innovative estimation framework that integrates Empirical Mode Decomposition (EMD) with a Gated Recurrent Unit (GRU) network enhanced by a self-attention mechanism (SAGRU). The EMD adaptively decomposes the raw battery capacity curve into three or more Intrinsic Mode Functions (IMFs) and a residual component. This process effectively separates high-frequency noise from the core degradation signal. These IMFs and the residual, which together represent the battery's aging characteristics, serve as the input for the SAGRU model. This allows the network to more efficiently learn the temporal features of battery aging. Furthermore, the integrated self-attention mechanism enables the model to weigh the significance of different features over time, improving focus on the most informative data points. To validate the model's performance, experiments were conducted on the NASA battery dataset, from which four batteries were selected. Data from three batteries constituted the training set, while the remaining battery was used as the test set to evaluate generalization capability. Results show that the proposed EMD-SAGRU model significantly outperforms conventional estimation methods in accuracy, providing a more robust solution for predicting SOH under diverse aging conditions.