Multiscale Attention Recurrent Network for Storage Estimation of Lithium-Ion Batteries

The state-of-health of batteries is crucial for ensuring their optimal performance. This paper presents a novel multiscale attention recurrent network, which is designed to enhance the accuracy and efficiency of state-of-health estimation in lithiumion batteries. Distinct from traditional models, this approach initially processes raw charge-discharge data across various time scales, employing feature fusion to more effectively capture temporal information. This method enhances model performance by introducing attention mechanisms following the residual network. The key innovation lies in the adept use of feature fusion for multiscale analysis of battery data, together with an attention mechanism to focus on informative features, which significantly enhances the accuracy of battery storage estimation. Research findings demonstrate that, based on historical operational data, this model can provide more precise and reliable predictions of battery health status, thereby significantly surpassing traditional methods.