重庆理工大学学报2025，Vol.39Issue(7)：59-66,8.DOI:10.3969/j.issn.1674-8425(z).2025.04.008
融合CEEMDAN分解与集成机器学习的锂电池剩余使用寿命预测方法
Lithium battery remaining useful life prediction method based on CEEMDAN decomposition and ensemble machine learning
摘要
Abstract
As a critical energy storage technology,lithium-ion power batteries are widely adopted on electric vehicles and portable devices.Discharge capacity is one of their key performance indicators.However,as usage time increases,a battery undergoes aging,and its discharge capacity exhibits nonlinear changes.To enhance the accuracy of battery state monitoring and improve the precision of remaining useful life(RUL)prediction,this paper proposes a novel method,which involves preprocessing battery discharge data through the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)algorithm,followed by modeling and predicting the preprocessed data using multiple machine learning models.First,the CEEMDAN algorithm is employed to preprocess the lithium-ion power battery's capacity aging data,breaking them down to constituent parts.By employing pearson correlation analysis and grid search methods,the decomposition parameters and the number of decomposition layers are determined.The process yields a residual data series and intrinsic mode functions(IMFs)data series.Then,a transformer neural network is employed to model and predict the obtained residual data series while a long short-term memory(LSTM)neural network is used to model and predict the IMFs data series.Finally,prediction results from these two models are fused to obtain the future capacity aging trajectory of the lithium-ion power batteries,from which the batteries' RUL is calculated.The method is verified by using NASA's lithium-ion batteries B0005 and B0018.Results demonstrate the lithium-ion power battery RUL prediction method exhibits superior robustness and better performance in nonlinear tracking.关键词
锂离子动力电池/剩余使用寿命/自适应噪声完全经验模态分解/长短时记忆神经网络Key words
lithium-ion power batteries/RUL/CEEMDAN/LSTM分类
信息技术与安全科学引用本文复制引用
张旭龙,周渝杰,张朝龙,杨忠..融合CEEMDAN分解与集成机器学习的锂电池剩余使用寿命预测方法[J].重庆理工大学学报,2025,39(7):59-66,8.基金项目
江苏省高等学校基础科学(自然科学)研究重大项目(23KJA480002) （自然科学）
