An inconsistency assessment method for backup battery packs based on time-series clustering

Slight differences in the production process and operation environment of individual cells in a battery pack result in inconsistencies from cell to cell, which become increasingly severe as the battery pack service time increases. Effectively assessing the inconsistency in a battery pack helps improve estimations of its state of charge, service condition, and degree of aging. However, it is difficult to recognize inconsistencies, especially for battery packs with low discharge frequencies, where there is no explicit parameter to directly measure inconsistency. Considering the available voltage as an evaluation factor, this paper proposes a method of inconsistency assessment for battery packs based on a clustering quality evaluation index that is applied to time-series data. First, a time-series model of an individual cell's voltage is created and the pattern distance is used to measure voltage differences between cells. Then, the k-medoids clustering algorithm is applied to realize each unit cell's cluster classification. Finally, the Davie–Bouldin clustering quality evaluation index is used to determine the inconsistency in the battery pack. To validate the proposed method, an example is demonstrated to determine the real capacity of individual cells. The results indicate that this method can accurately measure the inconsistency in a battery pack.