Because the data generated in the complex industrial manufacturing processes is multi-sourced and heterogeneous, it brings a challenge for addressing decision-making optimization problems embedded in the whole manufacturing processes. Especially, for inconsistent lithium-ion cell screening as such a special problem, it is a tough issue to fuse data from multiple sources in a lithium-ion cell manufacturing process to screen cells for relieving the inconsistency among cells in a battery pack with multiple cells configured in series, parallel, and series-parallel. This paper proposes a data-driven decision-making optimization approach (DDDMO) for inconsistent lithium-ion cell screening, which takes into account three dynamic characteristic curves of cells, thus ensuring that the screened cells have consistent electrochemical characteristics. The DDDMO method uses the convolutional auto-encoder to extract features from different characteristics curves of lithium-ion cells through multi-channels and then the features in different channels are combined into fusion features to build a feature base. It also proposes an effective sample generation approach for imbalanced learning using the conditional generative adversarial networks to enhance the feature base, thereby efficiently training a classifier for inconsistent lithium-ion cell screening. Finally, industrial applications verify the effectiveness of the proposed approach. The results show that the missing rate of inconsistent lithium-ion cells drops by an average of 93.74% compared to the screening performance in the single dynamic characteristic of cells, and the DDDMO approach has greater accuracy for screening cells at lower time costs than the existing methods.

由于在复杂的工业制造过程中生成的数据是多源且异构的，因此它对于解决嵌入整个制造过程中的决策优化问题提出了挑战。特别地，对于诸如这样的特殊问题的不一致的锂离子电池筛选，在锂离子电池制造过程中融合来自多个来源的数据以筛选电池以减轻具有多个电池的电池组中的电池间的不一致是一个棘手的问题。串联，并联和串并联配置。本文提出了一种基于数据驱动的决策优化方法（DDDMO），用于锂离子电池的不一致筛选，该方法考虑了电池的三个动态特性曲线，从而确保了筛选出的电池具有一致的电化学特性。DDDMO方法使用卷积自动编码器通过多通道从锂离子电池不同特征曲线中提取特征，然后将不同通道中的特征合并为融合特征，从而建立特征库。它还提出了一种有效的样本生成方法，用于使用条件生成对抗网络来增强特征库的不平衡学习，从而有效地训练用于锂离子电池筛选不一致的分类器。最后，工业应用验证了该方法的有效性。结果表明，与单个电池动态特性中的筛选性能相比，不一致的锂离子电池的丢失率平均下降了93.74％，