Accurate Li-ion battery (LiB) safety evaluation is crucial for reducing cell failures and promoting low-carbon economies. Existing AI anomaly detection methods lack validation in realistic battery settings due to complex failure mechanisms and limited real-world datasets. This paper presents a deep-learning framework for EV LiB anomaly detection using a dynamical autoencoder tailored for dynamical systems and informed by social and financial factors. Tested on a dataset of over 690,000 LiB charging snippets from 347 EVs, the model surpasses state-of-the-art methods, reducing expected EV battery fault and inspection costs. The research highlights deep learning's potential in improving LiB safety and the importance of social and financial information in model design.
