Capacity estimation of lithium-ion batteries with automatic feature extraction and graph-enhanced LSTM

Accurate battery capacity estimation is a key task in ensuring the safe and reliable operation of lithium-ion batteries and alleviating driver range anxiety. Most existing data-driven methods rely on manual feature extraction, requiring a significant amount of prior knowledge and mathematical computation, and feature extraction methods designed for a specific application may not generalize well to other scenarios. In this work, a feature extractor that combines the residual structure and attention mechanism is proposed to automatically extract effective features representing battery aging information without the need for manual intervention. In addition, this paper proposes a graph-enhanced LSTM model to make full use of the temporal and spatial information in the extracted feature maps for battery capacity estimation. Compared with other tested neural network models, the proposed model has higher accuracy on the MIT and Oxford datasets. The experimental results verify the effectiveness and feasibility of the proposed method.