Battery Capacity Estimation Based on Multi-health Features and Improved Long Short-Term Memory Neural Network

This is the authors' abstract. We don't add key points for this paper.

Accurately estimating the capacity of lithium-ion batteries is a key factor in ensuring safe and reliable battery operation. However, the estimation performance of data-driven methods can be affected by the selection of model hyperparameters. This paper proposes a model that uses an improved sparrow search algorithm to obtain the hyperparameters of the long short-term memory neural network to optimize the network for online battery capacity estimation. First, the health features are extracted from the battery partial charge and discharge data, and the model is trained offline with the features and capacity data of individual batteries. Subsequently, the multiple health features of the test batteries are input into the trained model, thus allowing the model to output the estimated capacity online. To assess model performance in diverse battery operating environments, experiments involve batteries at various temperatures and comparisons with different methods. Results demonstrate the proposed method's high accuracy, with a maximum estimation error of approximately 0.04 Ah.