Abstract:
The predominant method for identifying HSD cells entails prolonged static assessment of self-discharge efficiency, a process that is exceedingly time-intensive. This research presents a quick screening method utilizing T-statistics and support vector machine (TSM–SVM) for the efficient identification of batteries exhibiting HSD anomalies. The approach employs data gathered during the construction and aging phases of lithium-ion batteries, encompassing factors such as voltage and capacity. Essential parameters are delineated to define battery performance. A feature selection technique based on T-statistics is proposed to address aberrant screening conditions. Redundant features are eliminated by information entropy and the Pearson correlation coefficient, while T-statistics assess the sensitivity of each feature to internal micro-short circuits. To rectify class imbalance in the training set, the SMOTE algorithm is utilized to enhance the negative samples. The proposed TSM–SVM model is validated using fivefold cross-validation, continuously achieving an accuracy of about 96% and a 100% recall rate for the detection of HSD cells in all five validations.