基于长短期记忆网络-Transformer模型参数优化的锂离子电池剩余使用寿命预测

打开文本图片集

关键词：锂离子电池剩余使用寿命预测参数优化长短期记忆神经网络Transformer混合模型中图分类号：TM912；TP18 文献标志码：A DOI： 10.20104/j.cnki.1674-6546.20250107

RULPrediction ofLithium-Ion Battery Based onLSTM-Transformer Model Parameter Optimization

Gao Jianshu'，Hao Shiyu，DangYinuo （1.SchoolofTransportationScienceandEngineering，CivilAviation UniversityofChina，Tianjin3O3OO;2.Schoolof Electronic Information and Automation， Civil Aviation University ofChina，Tianjin 300300）

【Abstract】Inorder to further enhance prediction accuracy，this paper proposes a lithium-ion batery Remaining UsefulLife （RUL）predictionmethod basedonparameteroptimizationof theLSTMnetwork-Transformer model.This methodselectsmodel hyperparametersusing thegridsearch technique，employsLong Short-Term Memory （LSTM） network to extractlongshort-termdependenciesfromlithium-ionbaterytimeseriesdata，utilizes Transformer forselfatentionmechanismprocessngofglobalinformationandoptimizationofhyperparameters，andperformsfinal life predictionthroughafullconnectedlayer.Experimental validationusing NationalAeronauticsandSpaceAdministration （NASA）and Center for AdvancedLife Cycle Enginering （CALCE）datasets demonstrate that themodel outperforms standaloneLSTM，Transformer，andotherneural network modelsacross multipleevaluationmetricssuchasshorter sequencelengths，fewer hidenlayers，and fewer training iterations，exhibiting higher predictionaccuracyandrobustness. Furthermore，comparative experiments withdiferent bateries furthervalidate the model’sgeneralization capabilityacross diverse battery datasets.

KeyWords：Lithium-ion battery，RUL prediction，Parameter optimization，LSTM neural network，Transformer，Hybridmodel

【引用格式】高建树，郝世宇，党一诺.基于长短期记忆网络-Transformer模型参数优化的锂离子电池剩余使用寿命预 测[J].汽车工程师，2026（1）：32-39. GAOJS，HAO SY，DANGYN.RULPredictionof Lithium-Ion Battery BasedonLSTM-Transformer Model ParameterOptimization[J].Automotive Engineer，2026（1）：32-39.

1前言

锂离子电池剩余使用寿命（RemainingUsefulLife，RUL）的准确预测不仅有助于提高设备的可靠性，还能降低维护成本[1-3]

传统的寿命预测方法往往依赖于统计模型或简单的机器学习算法，然而，这些方法在处理复杂的时间序列数据时存在局限性。（剩余9270字）