An enhanced sorting method for retired battery with feature selection and multiple clustering

Abstract

Considering the promising prospects of retired power batteries in second-life utilization, it becomes imperative to enhance their consistency through a reasonable sorting procedure. However, the selection of input features and clustering algorithms significantly affects the performance of the battery sorting. Thus, an enhanced sorting method with feature selection and multiple clustering is proposed to enable a reliable sorting of the retired batteries. To prioritize the importance of features, the Pearson correlation coefficient and gird search are employed to identify features with the highest correlation to capacity. Additionally, a scoring fusion mechanism is proposed on the foundation of the improved K-mean algorithm, the fuzzy C-mean algorithm, the Gaussian hybrid algorithm, the hierarchical clustering, and the spectral clustering. With the well-designed fusion mechanism, the integration of sorting methods across different battery capacity distributions is reinforced. The validation experiments on 97 retired cells and up to 600 simulated cells have demonstrated the high classification accuracy of the proposed method on diverse tests.