Principle component analysis-based optimized feature extraction merged with nonlinear regression model for improved state-of-health prediction

The state-of-health (SOH) estimation and prediction is critical for battery energy storage systems (BESS) to detect poor battery performance. The BESS consists of a high-energy battery pack with series and parallel connections. Unlike unit cells, battery packs with series and parallel combinations must account for cell-to-cell imbalance. The cell-to-cell imbalance indicates deviations in voltage, state-of-charge (SOC), and temperature. A serious imbalance in battery packs can hamper their capacity, power, and efficiency. However, conventional methods of SOH estimation and prediction only consider one or two degradation features. To improve the SOH prediction performance of a battery pack, more features than those used for a unit cell must be considered. In this study, an optimal regression model is used to propose a feature extraction method for reflecting new degradation features. Feature extraction based on principal component analysis takes into account various degradation features. The proposed method can highlight various BESS degradation features and improve SOH prediction performance through combination of the principal component analysis and optimal regression model.