关键词: 锂离子电池, 卷积神经网络, 不一致性, 等效电路模型, SOH评估及分筛

Abstract:

With the rapid development of electric vehicles, ensuring the health and performance evaluation of large-scale battery systems has become a crucial technological challenge. This paper focuses on addressing this issue by constructing an equivalent circuit model for battery packs. The health status of multiple battery packs is simulated using a combination of orthogonal parameters, representing different types of inconsistencies. Additionally, a dataset of battery pack samples is generated through model simulations. To classify the battery packs efficiently, a convolutional neural network (CNN) model is established. This model extracts morphological features from the local charging voltage curve images of the battery packs. These features serve as input for the CNN, enabling quick classification. Four parameters, namely available capacity, available energy, capacity utilization rate, and energy utilization rate, are selected for evaluation. The weight of each parameter is determined using the Analytic Hierarchy Process. Consequently, a comprehensive health evaluation index is proposed, which takes into account various performance characteristics of battery packs. This index facilitates the screening of battery components based on their evaluation scores. The classification model is trained and tested using the simulated dataset, yielding promising results. The constructed battery pack classification model achieves an accuracy of over 97% on the test set. The effectiveness of this method is further validated by evaluating the model with various indexes, such as the confusion matrix for classification tasks. Overall, the proposed method, which involves the visual feature extraction-based comprehensive assessment and screening of battery pack health status, contributes to the advancement of battery performance evaluation research. It also provides a new theoretical foundation for the health supervision of battery systems.

Key words: lithium-ion batteries, convolutional neural network, cell inconsistency, equivalent circuit model, SOH assessment screening