应用支持向量机的锂电池不可逆析锂检测研究

The fast charging technology of lithium-ion batteries is crucial for enhancing the endurance of batteries in electric vehicles, electrochemical storage, and mobile terminal products. However, issues like lithium plating resulting from fast charging can rapidly decline battery capacity and cause safety incidents like internal short circuits. Therefore, research on real-time and effective lithium plating detection schemes is crucial to overcome the current bottleneck in fast-charging technology. Traditional lithium plating detection methods usually involve destructive testing using expensive precision instruments, providing relatively accurate results only in laboratory environments. Recently, some methods have been proposed for detecting reversible lithium plating in batteries, but more research is needed on the more dangerous irreversible lithium plating. Therefore, this paper proposes a support vector machine-based method to achieve real-time, in-situ detection of lithium plating status by inputting the required features into the model. Firstly, small-rate discharge increment capacity curves for different aging states and operating conditions are obtained according to the data logger, including voltage, current, and capacity. Secondly, features of the capacity increment curvealong with the absolute capacity of the battery, such as peak value, peak voltage, and peak area, are extracted to determine the current lithium plating status after disassembling the battery. This process facilitates the construction of the dataset required for the model. Thirdly, correlation analysis is conducted to select parameters strongly correlated with lithium plating and effectively reflecting changes in battery performance. Finally, a nonlinear support vector machine is used as a binary classification algorithm to identify the battery's lithium plating status by inputting the selected features into the trained model. Comparisons indicate that the proposed classification algorithm achieves higher accuracy in detecting irreversible lithium plating in lithium batteries than other commonly used binary machine learning algorithms, reaching 94.2%. Meanwhile, selecting 10 features as model inputs yields the optimal results, with a lithium battery irreversible lithium plating detection rate of 95.2% and a false detection rate of 8%, which is important for avoiding model overfitting problems and enhancing model versatility. Actual accuracy validation results demonstrate that the proposed method for detecting irreversible lithium plating exhibits good accuracy under different aging states and charging strategy conditions. With the increase in detection samples, the lithium plating detection rate can reach a high accuracy of 95%, while the false detection rate is less than 10%. Conclusions drawn from validation analysis include: (1) The proposed model significantly improves detection accuracy and effectively avoids the issue of unclear manual threshold settings. (2) The proposed model only requires voltage, current, and capacity data. It exhibits rapid, non-destructive, in-situ detection of irreversible lithium plating in practical applications, which is more practical than traditional lithium plating detection methods. (3) The method establishes the correlation between changes in capacity increment curve features and irreversible lithium plating. The model is trained using support vector machines to accuratelyidentify the battery's lithium plating status.