Accurate life prediction of lithium-ion batteries is essential for the safety and reliability of smart electronic devices, and data-driven methods are one of the mainstream methods nowadays. However, existing prediction methods suffer from the problems such as lack of practical meaning of features and insufficient interpretability. To address this problem, this article proposes a battery degradation and capacity prediction model based on the Granger causality (GC) test and the long short-term memory network. First, initial health indicators are set from the monitoring data of the battery. Second, the vector autoregressive model and the GC test are used to select causal features that are associated with capacity degradation. Then, the impulse response analysis approach is proposed for the first time to analyze the exact influence of the features on capacity degradation and combine the battery aging mechanism, further clarifying the interpretability of the selected features. Finally, using the causal features as model input, a prediction model based on the long short-term memory network is constructed. The experimental results of the two datasets show that the minimum root mean square error is 0.0093 Ah and 0.9635 mAh with the mean relative errors of 0.25% and 0.13%, which verifies the validity and accuracy of the proposed method.

中文翻译：

---

考虑因果特征的锂离子电池退化和容量预测模型


锂离子电池的准确寿命预测对于智能电子设备的安全性和可靠性至关重要，而数据驱动方法是当今的主流方法之一。然而，现有的预测方法存在特征缺乏实际意义、可解释性不足等问题。针对这一问题，本文提出了一种基于格兰杰因果关系（GC）测试和长短期记忆网络的电池退化和容量预测模型。首先，根据电池的监测数据设置初始健康指标。其次，使用向量自回归模型和 GC 测试来选择与容量退化相关的因果特征。然后，首次提出脉冲响应分析方法来分析特征对容量衰减的确切影响，并结合电池老化机制，进一步阐明所选特征的可解释性。最后，以因果特征作为模型输入，构建基于长短期记忆网络的预测模型。两个数据集的实验结果表明，最小均方根误差分别为0.0093 Ah和0.9635 mAh，平均相对误差分别为0.25%和0.13%，验证了该方法的有效性和准确性。