Assessment of the battery pack consistency using a heuristic-based ensemble clustering framework

With the deterioration of the cells' consistency, the overall performance and maintenance of the battery energy storage system (BESS) is significantly limited. In this thread, assessing the battery pack consistency is always critical to manage the BESS operation. Since the real-world BESS lacks the opportunity to receive a trustworthy label, it's troublesome to accurately evaluate the consistency of a battery pack. Thus, this paper proposes a novel heuristic-based ensemble clustering framework enabling to evaluate the consistency of the battery pack according to the statistical consistency indicators (CIs) from the daily operation measurement data of BESS. An automatic formulation procedure is designed to intelligently select the useful CIs and effective clustering algorithms, where an enhanced genetic algorithm is used to optimize the ensemble clustering model simultaneously. Twelve CIs accessible from practical applications are chosen to fully use the voltage and temperature information. The validation of the proposed method is proved on datasets from constructed battery packs and real-world BESS. The findings reveal that, across both datasets, the average root mean square error (RMSE), mean absolute error (MAE), and r-square (R²) values for the assessments of normalized battery pack consistency are 8.54 %, 6.96 %, and 0.91, respectively.