Reliable and accurate state of charge (SOC) estimation is pivotal to enhancing the performance and safety of lithium-ion batteries for battery management system. Extended Kalman filter (EKF) and smooth variable structure filter (SVSF) has alreadly been used for estimating the SOC of lithium-ion batteries. However, either of the two methods only considers the accuracy or robustness of SOC estimation, and cannot give attention to both aspects simultaneously. To this end, this paper proposes a particle swarm optimized combined method that leverages their complementary strengths fully, enhancing the accuracy and robustness of SOC estimation. The effectiveness and superiority of the proposed method are verified by battery test experiments conducted at different ambient temperatures and battery aging states, and comparing to EKF, SVSF, and manual tuning combined method. Experimental results demonstrate that the root mean square error of the proposed method is kept below 6% across all conditions, highlighting its application potential for SOC estimation in complex and variable operational conditions.

Sustainable development in the energy storage sector requires batteries to be graded before leaving the factory so that their performance matches the requirements of intended applications, thereby maximizing service potential and minimizing resource waste caused by premature retirement. However, conventional grading relies primarily on initial performance and does not account for the future capacity degradation trajectory or service life, and limited early-cycle data are typically available. This paper therefore proposes a method that uses only two early aging cycles to estimate both the capacity degradation trajectory and the remaining useful life (RUL). The proposed method, termed CNN-LSTM-SE, combines a convolutional neural network-long short-term memory (CNN-LSTM) architecture with a stretched exponential (SE) model. The network takes the resampled voltage-capacity curves obtained during the constant-current charging segments of the first and third cycles as input and estimates the parameters of the SE model. The predicted parameters are then used to reconstruct the capacity degradation trajectory, from which the battery RUL is estimated. Experimental results show that the proposed method achieves mean absolute percentage errors (MAPE) of 5.06% and 0.73% for RUL estimation and capacity degradation trajectory prediction, respectively, with corresponding root mean square errors (RMSEs) of 4.54 cycles and 19.96 mAh. Compared with TCN-SE, RF-SE, SVR-SE, and GPR-SE, the proposed method achieves the lowest MAPE and RMSE for both RUL estimation and degradation trajectory prediction. Overall, the proposed method enables accurate estimation of the degradation trajectory and RUL using only two early cycles, showing promise for sustainable battery deployment.

Large-scale deployment of renewable energy increases the demand for reliable SOH estimation of lithium-ion battery energy storage systems, while intermittent renewable generation and flexible power dispatch result in fragmented and stochastic charging segments, significantly complicating SOH estimation. To address this issue, an SOH estimation method is proposed that constructs capacity increment sequences from partial charging segments with stochastic starting positions, extracts degradation-related statistical features including standard deviation, maximum value, skewness, and kurtosis, and maps these features to SOH using a convolutional neural network-bidirectional gated recurrent unit (CNN-BiGRU) model. Experimental results show that the proposed method achieves the best overall performance on the CALCE dataset, where the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error are 0.34%, 0.46%, and 0.39%, respectively, and maintains the best performance among the compared methods on the NASA dataset. Further analysis shows that the proposed method achieves accurate SOH estimation across all segment positions, with MAE and RMSE consistently below 0.5% and 0.7%, respectively, without reliance on a specific starting position, while segment lengths within the range of 30-50 provide sufficient health-related information for stable estimation under random partial charging conditions.