The proposed DOA-VMD-CNN-TimeXer framework for RUL-oriented prognosis through capacity trajectory prediction and reconstruction shows closer agreement with the measured degradation trajectories and more stable prediction behavior than the compared methods, indicating its effectiveness for lithium-ion battery RUL-oriented prognosis under the evaluated CALCE settings.
Reliable state of health (SOH) estimation plays an important role in the safe and stable operation of lithium-ion battery energy storage systems. Nevertheless, the nonlinear degradation characteristics and complex aging behaviors of batteries hinder accurate SOH estimation. This study proposes a Whale Optimization Algorithm (WOA)-optimized Convolutional Neural Network (CNN)–Temporal Convolutional Network (TCN)–Temporal Pattern Attention (TPA) framework for lithium-ion battery SOH estimation. Multiple health factors are extracted from charge–discharge curves to characterize battery degradation behaviors. Neighborhood-based imputation and Hampel-Median Absolute Deviation (MAD) correction are employed to handle missing values and local outliers, while Pearson correlation analysis is applied to evaluate the relevance of extracted features. CNN module captures local degradation patterns, TCN module learns long-term aging dependencies, and TPA mechanism enhances the representation of critical degradation stages. Furthermore, WOA adaptively optimizes key hyperparameters to improve model performance and robustness. The proposed framework is validated using NASA and CALCE battery datasets. Experimental results demonstrate that, compared with the CNN-TCN model, the proposed method reduces RMSE by 7.49–50.81% on NASA datasets and 15.83–60.38% on CALCE datasets, achieving higher estimation accuracy and stability.
State of charge (SOC) is one of the most critical state variables in lithium-ion batteries for energy storage systems and electric vehicles. SOC estimation under varying temperatures and operating conditions remains challenging due to the difficulty of simultaneously representing local dynamics, long-range temporal dependencies, and salient information. To this end, this study proposes a hybrid CNN–Mamba–Attention network. A convolution-based feature extraction and residual fusion module is introduced ahead of the Mamba backbone to enhance local variation information in the input signals and to facilitate the modelling of long-term temporal dependencies during battery operation. To further refine the learned representation, an attention mechanism is applied after the Mamba module, allowing important features to be highlighted for more accurate SOC estimation. The proposed method is validated using publicly available dynamic test data for lithium-ion batteries under three temperature settings and three operating conditions. Experimental results show that the model achieves a minimum root mean square error (RMSE) of 0.75%, a lowest mean absolute error (MAE) of 0.59%, and a maximum coefficient of determination (R²) of 0.9993, demonstrating strong generalization capability and robustness across different working conditions.
Accurate remaining useful life (RUL) prediction is essential for condition-based maintenance and safe aero-engine operation. To address non-stationary monitoring data, limited informative degradation samples, and the difficulty of jointly modeling local degradation patterns and temporal dependencies, this study proposes a degradation-aware dynamic masking augmentation method combined with a multiscale CNN–Transformer network. The 21 sensor variables in the NASA C-MAPSS dataset are first grouped by physical meaning and reconstructed into four-dimensional state features. A degradation-state score integrating local variance and trend slope is then used to adapt temporal masking probabilities across degradation stages, while feature masking probabilities are assigned according to feature importance. Masked positions are filled with adjacent unmasked observations, and invalid augmented samples are removed through trend consistency verification. A multiscale CNN with channel attention extracts local degradation features, and a Transformer encoder captures temporal dependencies. Bayesian optimization is used to determine key hyperparameters. On FD001, the proposed method achieves an MSE of 348.3970, an MAE of 8.3908, and an R2 of 0.9227, reducing MSE and MAE by 4.27% and 28.42%, respectively, compared with ML-RFR. It also achieves the highest R2 of 0.9369 on FD003. Cross-dataset and multi-seed experiments further confirm its applicability and stability.
Accurate State-of-Health (SOH) prediction is essential for deploying retired electric vehicle batteries into reliable second-life energy storage systems. However, this task is challenged by sparse and noisy operational data from onboard Battery Management Systems (BMS). This study systematically evaluates a spectrum of deep learning architectures for SOH forecasting under BMS-style data constraints derived from laboratory cycling data: a BiLSTM on aggregated cycle statistics (Model A), preliminary zero-shot transfer to a single unseen cell (Model B), a waveform BiLSTM with full intra-cycle voltage, current, and temperature trajectories (Model C), a baseline TCN (Model D) and a probabilistic TCN-GPR hybrid (Model E). All models are constrained to identical low-fidelity BMS-style variables extracted from the NASA battery aging dataset. Model C achieves the lowest point accuracy error of 0.46% ± 0.18% MAE across five random seeds, demonstrating that high-resolution waveform inputs capture degradation signatures, notably voltage plateau morphology, transient dynamics, and implicit SOC information, that aggregated features irreversibly lose. Model D using the same waveform inputs and evaluation protocol as Model C, achieves a MAE of 2.99% at a single seed, providing direct architectural comparison evidence that the BiLSTM’s position-sensitive temporal summarization outperforms GlobalAveragePooling1D under these conditions. Model E achieves a higher MAE of 2.12% ± 0.33% but uniquely provides calibrated predictive distributions of 99.4% ± 1.2% coverage, NLL = −1.877 ± 0.038, with approximately uniform 95% predictive intervals (mean width 19.83% SOH across 34 test cycles at seed = 42), reflecting the near-constant posterior variance produced by the large optimized GPR length-scale under the frozen two-stage training design. A paired t-test confirms that Model C statistically significantly outperforms Model E on point accuracy (p < 0.01). Isotonic regression recalibration reduces mean calibration error from 0.138 to 0.010, demonstrating that shape-level miscalibration is correctable post hoc. The central implication for second-life battery deployment is a clear accuracy–uncertainty trade-off: Model C is preferred when point estimates suffice, while Model E is essential for risk-aware decisions requiring confidence intervals.
Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is essential for the reliability and safety of modern energy systems. However, the capacity regeneration phenomenon, a temporary recovery of capacity during cycling or rest, introduces non-monotonic fluctuations in degradation trajectories, posing significant challenges to existing prediction models.Many current approaches treat CR as noise or overlook its physical significance, limiting interpretability and accuracy. To address this, we propose a hybrid framework that explicitly models both regenerative and degenerative battery behaviors. The method uses Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to decompose capacity sequences into high-and low-frequency components. A Graph Convolutional Network-Long Short-Term Memory (GCN-LSTM) branch captures dynamic regeneration features, while a Deep Neural Network (DNN) branch learns long-term degradation trends. These are fused to reconstruct the full degradation path and predict RUL. The CEEMDAN-GCN-LSTM-DNN hybrid model achieves a mean absolute percentage error below 0.15%, outperforming several state-of-the-art baselines. It also demonstrates strong robustness under data-limited conditions and effectively captures complex capacity regeneration patterns often missed by conventional methods. This study offers a new perspective for handling non-monotonic battery degradation and provides a useful tool for battery health management and predictive maintenance.