Applied Sciences · 2025 · 30 citations · 34 references
ComputingComputer ScienceData ScienceArtificial IntelligenceMachine LearningTime Series AnalysisAttention NetworksBattery SystemsBattery TechnologyDeep LearningNeural NetworksPredictive Health ModelingBattery Aging ModelsBattery Degradation ModelingBattery ModelingBattery PrognosticsConvolutional Neural NetworksDegradationLithium-ion BatteriesState-of-health Estimation
Lithium-ion batteries play a vital role in human society. Therefore, it is of critical significance to reliably predict the evolution of State of Health (SOH) degradation patterns in order to improve the high accuracy and stability of lithium-ion battery SOH prediction. This paper proposes a novel SOH predication method by combing the four-vector intelligent metaheuristic (FVIM) with the CNN-LSTM-Attention basic model. The model adopts the collaborative architecture of a convolutional neural network and time series module, strengthens the cross-level feature interaction by introducing a multi-level attention mechanism, then uses the FVIM optimization algorithm to optimize the key parameters to realize the overall model architecture. By analyzing the charging voltage curve of lithium-ion batteries, the health factors with high correlation are extracted, and the correlation between the health factors and battery capacity is verified using two correlation coefficients. After the model is verified on a single NASA battery aging dataset, the model is compared with other models under the same relevant parameters and environmental settings to verify the high-precision prediction of the model. During the analysis and comparison process, CNN-LSTM-Attention-FVIM achieved a high fitting ability for battery SOH prediction estimation, with the mean absolute error (MAE) and root mean square error (RMSE) within 0.99% and 1.33%, respectively, reflecting the model’s high generalization ability and high prediction performance.
34
Manh‐Kien Tran, Manoj Mathew, Stefan Janhunen et al. · Journal of Energy Storage · 2021 · 510 citations
Engineering, Chemical Engineering, Electrical Engineering +16