Lithium Battery State of health Prediction Based on Automatic Feature Selection

Aiming at the issues of data loss caused by sensor problems and the impact of manual feature extraction on real-time performance in the online assessment of lithium battery State of health (SOH), this paper proposes a lithium battery SOH prediction method based on automatic feature selection. The method first employs fixed time windows and polynomial fitting techniques to enhance data, repairing missing values and improving data quality. Subsequently, an automatic feature selection module is designed, which performs initial feature extraction through a simple convolutional layer and constructs graph structures for data from different time windows using three different aggregation methods to fully extract local important features and global structural information. Finally, the extracted features are transformed through a linear layer and input into the Informer prediction model to obtain the lithium battery SOH prediction results. The effectiveness and superiority of the proposed method are validated through simulation tests on public datasets.