机械滥用场景下锂离子电池温度多模态神经网络预测方法

打开文本图片集

关键词：锂离子电池；机械滥用；注意力机制；温度预测

中图分类号：TM734文献标志码：A

DOI：10.7652/xjtuxb202602004 文章编号：0253-987X（2026）02-0038-11

Multimodal Neural Network-Based Temperature Prediction Method for Lithium-Ion Batteries Under Mechanical Abuse Scenarios

LI Jie，LI Qian，QIN Zhengpeng，HUANG Xinrong （School of Energy and Electrical Engineering，Chang'an University，Xi'an 71oo64，China）

Abstract：To address the issue of low temperature prediction accuracy in lithium-ion batteries under mechanical abuse scenarios caused by single-modal modeling，simplistic fusion strategies， and insufficient physical constraints， a maximum temperature prediction method based on multimodal neural networks is proposed. First，1865O-type lithium-ion batteries with a capacity of 1 200mA⋅h were selected，and mechanical compression experiments were conducted within a state of charge（SOC） range of 10% to 90% . Multi-source data including thermal images，SOC， voltage，load，and deformation were collected to construct a multimodal dataset comprising over 200 sample groups. Subsequently，multiscale convolutional modules，attention mechanisms，and bidirectional long short-term memory （Bi-LSTM） networks were used to extract spatiotemporal features from thermal images，while one-dimensional convolutional neural networks combined with Bi-LSTM were employed to extract electrochemical and mechanical spatial-temporal features. Cross-modal adaptive feature fusion was achieved through a transformer fusion mechanism. Finaly，a physics-informed loss function incorporating temperature change rate constraints was introduced to improve prediction rationality and robustness.The results show that the multimodal neural network model achieves coefficients of determination of O.972 to O.990 for short-term predictions （1—3 steps） and remains stable above O.9oo for medium-to long-term predictions （6—15 steps）. The model enables effective multi-step temperature prediction under mechanical abuse conditions and demonstrates significant accuracy advantages and good adaptability across different SOC levels and prediction horizons， providing technical support for multi-level early warning and safety management of lithium-ion battery thermal runaway.

Keywords： lithium-ion battery；mechanical abuse;attention mechanism；temperature prediction

随着电动汽车与储能系统的快速发展，锂离子电池因高能量密度和长寿命而被广泛应用，但其在机械滥用条件下的安全性仍是关键制约因素[1-2]机械冲击或挤压可能导致内部结构损伤、隔膜破裂和电解液泄漏，引发短路与热失控进而产生高温，严重威胁设备与人员安全[3]。（剩余14356字）