Abstract: The graded utilization of power batteries relies on accurate state of health (SOH) evaluation. However, existing methods suffer from limitations such as dependence on single features and neglect of temporal dynamic changes. This paper proposes a hybrid convolutional neural network-long short-term memory (CNN-LSTM) model based on industrial computed tomography (CT) images. The method extracts three-dimensional internal structural features of battery cells from industrial CT images, and the CNN-LSTM hybrid model simultaneously captures spatial features and temporal degradation patterns to achieve accurate SOH prediction. Experimental results demonstrate that the proposed method achieves a mean squared error (MSE) of 0.033. Compared to traditional internal resistance methods and electrochemical impedance spectroscopy (EIS) methods, the mean absolute error (MAE) is reduced by over 30% (from 0.038 to 0.026). Moreover, the prediction error rate remains below 4.2% across different SOH ranges (40%～100%). This study provides a highprecision tool for screening batteries in graded utilization and promotes the sustainable development of the new energy industry.

Key words: graded utilization; battery state of health evaluation; convolutional neural network; long short-term memory; industrial CT

摘要： 动力电池梯次利用依赖精准的电池健康状态（SOH）评价，但现有方法存在依赖单一 特征、忽略时序动态变化等问题。文章提出一种基于工业计算机断层扫描（CT）影像的卷积 神经网络-长短期记忆网络（CNN-LSTM）融合模型，通过工业 CT 影像提取电芯内部三维结 构特征，结合 CNN-LSTM 融合模型同步建模空间特征与时序退化规律，实现 SOH 的精准预 测。实验表明，该方法的均方误差（MSE）为 0.033，较传统内阻法和电化学阻抗谱（EIS） 法，平均绝对误差（MAE）低超 30%（从 0.038 降至 0.026），且在不同 SOH 区间（40%～100%） 的预测误差率均低于 4.2%。该研究为梯次利用电池筛选提供了高精度工具，并推动新能源产 业的可持续发展。

关键词: 梯次利用；电池健康状态评价；卷积神经网络；长短期记忆网络；工业 CT